ChatGPT Fanout Explorer

Version 2026-09-22.8, released 22 September 2026. What changed

A bookmarklet that shows every search ChatGPT runs behind the scenes for a chat, live as it answers or on a saved chat, with the pages each search got back and which of them were cited. Read-only.

What it shows

Live and saved chats. On a new chat it fills as ChatGPT searches, a batch of several searches at a time, then settles once the answer is done. On a chat from your history it shows the whole fan-out at once. For every search it gives you the pages that came back and which of them were cited. Since September 2026 ChatGPT no longer saves the query text with the chat, so the bookmarklet captures it from the answer as it streams: Click the bookmark before you send the prompt and the queries are kept for that chat. The freshness window and the site limit are no longer sent at all, so those columns are empty on new chats and still filled on chats saved before the change.

Install

Recommended: Drag this button to your bookmarks bar (Cmd+Shift+B on a Mac or Ctrl+Shift+B on Windows shows the bar). It is short, so it survives any drag or paste and fetches the latest code from GitHub each time you click it:

ChatGPT Fanout Explorer

Or copy this and paste it as the URL of a new bookmark (bookmark any page, right-click it, Edit):

javascript:(()=>{var d=document,n=(d.querySelector('script[nonce]')||{}).nonce||'',s=d.createElement('script');if(n)s.nonce=n;s.src='https://cdn.jsdelivr.net/gh/oritsimu-new/fanout-explorer@main/chatgpt-fanout-explorer.min.js?t='+Math.floor(Date.now()/36e5);s.onerror=()=>{var b=d.createElement('div');b.style.cssText='position:fixed;top:12px;right:12px;z-index:2147483647;background:rgb(17,17,17);color:rgb(235,235,235);font:13px/1.4 -apple-system,Segoe UI,Helvetica,Arial,sans-serif;border:1px solid rgb(70,70,70);border-radius:10px;padding:12px 14px;max-width:420px';b.textContent='ChatGPT Fanout Explorer could not load its code from GitHub. Check your connection, or use the standalone version on the install page.';b.onclick=()=>b.remove();d.body.appendChild(b)};d.head.appendChild(s)})();

Standalone version

The same tool with all of its code inside the bookmark, so it never contacts GitHub. It is about 50KB, which some drags and pastes on Windows do not survive, so use the copy button and the paste route rather than dragging:

ChatGPT Fanout Explorer (standalone)

javascript:(async()=>{const e=['fast','slow','image','product','news','video','finance','weather','sports','business'],t=['open','find','click','length','screenshot','scroll'],n=['fast','slow','news','product','video','finance','weather','sports','web'],a='2026-09-22.8',o=['business','image'],s=[],r=[],i=[],c={id:'',title:'',at:'',prompts:0,promptList:[],fetched:0,cited:0,redditFetched:0,redditCited:0,unknownTurns:0,hiddenQueries:0,liveRounds:0,savedRounds:0},d=e=>{try{return new URL(e).hostname.replace(/^www\./,'').toLowerCase()}catch(e){return''}},h=e=>String(e||'').toLowerCase().replace(/^https?:\/\//,'').replace(/^www\./,'').split(/[?\x23]/)[0].replace(/\/$/,''),l=(e,t)=>e===t||e.endsWith('.'+t),p='fo-export:q:v1',u=()=>{try{return JSON.parse(localStorage.getItem(p)||'{}')||{}}catch(e){return{}}},m=e=>{if(!e)return null;if(e.search_model_queries&&Array.isArray(e.search_model_queries.queries)){const t=e.search_model_queries.queries.filter(e=>'string'==typeof e&&e.trim()).map(e=>e.trim());return t.length?{q:t,t:[]}:null}if(Array.isArray(e.search_queries)&&e.search_queries.length){const t=[],n=[];return e.search_queries.forEach(e=>{e&&'string'==typeof e.q&&e.q.trim()&&(t.push(e.q.trim()),n.push('string'==typeof e.type?e.type:''))}),t.length?{q:t,t:n}:null}return null},g=(e,t,n,a)=>{if(!e||!a)return;const o=u(),s=o[e]||(o[e]={});s._at=Date.now();const r={q:a.q,t:a.t,msg:n||'',at:Date.now()};if(t&&(s[t]=r),n&&(s['@'+n]=r),t||n||(s['n'+Object.keys(s).length]=r),(e=>{try{Object.keys(e).sort((t,n)=>(e[n]._at||0)-(e[t]._at||0)).slice(40).forEach(t=>{delete e[t]}),localStorage.setItem(p,JSON.stringify(e))}catch(e){}})(o),window.__foTap&&window.__foTap.onCapture)try{window.__foTap.onCapture(e)}catch(e){}};if(!window.__foTap){const e=window.__foTap={onCapture:null,captured:0,streams:0},t=window.fetch,n=()=>(location.pathname.split('/c/')[1]||'').split(/[?\x23]/)[0],a=(t,a)=>{const o=t&&t.v,s=t&&t.conversation_id||o&&o.conversation_id||a.cid||n();s&&(a.cid=s);const r=o&&o.message;if(r&&r.id){a.lastMsg=r.id,a.lastParent=r.metadata&&r.metadata.parent_id||'';const t=m(r.metadata);return void(t&&(e.captured++,g(s,a.lastParent,r.id,t)))}(t&&'patch'===t.o&&Array.isArray(o)?o:t&&t.p?[t]:[]).forEach(t=>{if(!t||'string'!=typeof t.p)return;let n=null;/search_model_queries$/.test(t.p)?n=m({search_model_queries:t.v}):/search_queries$/.test(t.p)&&(n=m({search_queries:t.v})),n&&(e.captured++,g(s,a.lastParent,a.lastMsg,n))})},o=(e,t)=>{let n;for(t.buf+=e;(n=t.buf.indexOf('\n\n'))>=0;){const e=t.buf.slice(0,n);t.buf=t.buf.slice(n+2);const o=e.match(/^data:\s?(.*)$/m);if(!o)continue;let s;try{s=JSON.parse(o[1])}catch(e){continue}if(s&&'object'==typeof s)try{a(s,t)}catch(e){}}};window.fetch=async function(n,a){const s=await t.apply(this,arguments);try{const t='string'==typeof n?n:n&&n.url||'',r=String(a&&a.method||n&&n.method||'GET').toUpperCase(),i=s.headers&&s.headers.get('content-type')||'';if('POST'===r&&/\/backend-api\/(f\/)?conversation(\?|$)/.test(t)&&/event-stream/.test(i)&&s.body){const t=s.body.tee(),n={buf:'',cid:'',lastMsg:'',lastParent:''};return e.streams++,(async()=>{const e=t[1].getReader(),a=new TextDecoder;try{for(;;){const t=await e.read();if(t.done)break;o(a.decode(t.value,{stream:!0}),n)}}catch(e){}})(),new Response(t[0],{status:s.status,statusText:s.statusText,headers:s.headers})}}catch(e){}return s}}const f=a=>{s.length=0,r.length=0,i.length=0,c.title='string'==typeof a.title?a.title:'',c.fetched=0,c.cited=0,c.redditFetched=0,c.redditCited=0,c.unknownTurns=0,c.hiddenQueries=0,c.liveRounds=0,c.savedRounds=0,c.promptList=[];let p='',g=0,f=0,w=null,y='';const b=(e=>{const t=[];let n=e.current_node;const a=new Set;for(;n&&e.mapping[n]&&!a.has(n);)a.add(n),t.push(e.mapping[n]),n=e.mapping[n].parent;return t.reverse()})(a),x=u()[a.conversation_id||c.id]||{},C={};b.forEach(e=>{const t=e.message;if(!t)return;const n=m(t.metadata);if(!n)return;const a=t.metadata&&t.metadata.parent_id||e.parent||'';a&&(C[a]=n),C['@'+t.id]=n});const k=(e,t,n)=>{if(e[t])return e[t];for(let t=n+1;t<b.length;t++){const n=b[t].message;if(n&&n.author){if('user'===n.author.role||'web.run'===n.recipient||'web'===n.recipient)break;if(e['@'+n.id])return e['@'+n.id]}}return null},v={},T=(e,t,n)=>{if(!t||!t.url)return;const a=h(t.url),o='string'==typeof t.title?t.title.slice(0,160):'',s='number'==typeof t.pub_date&&t.pub_date>0?t.pub_date:0,r='string'==typeof t.snippet?t.snippet.replace(/\s+/g,' ').trim().slice(0,240):'',i=e.seen.get(a);if(i)return!i.title&&o&&(i.title=o),!i.date&&s&&(i.date=s),void(!i.snippet&&r&&(i.snippet=r));const c=t.ref_id||{},l={url:a,raw:String(t.url),host:d(t.url),key:e.turn+'|'+n+'|'+c.ref_type+'|'+c.ref_index,cited:!1,title:o,date:s,snippet:r};e.entries.push(l),e.seen.set(a,l)};b.forEach((n,a)=>{const o=n.message;if(!o||!o.author)return;const d=o.author.role,l=o.metadata||{};if('user'===d&&o.content&&o.content.parts)return p=o.content.parts.filter(e=>'string'==typeof e).join(' ').replace(/\s+/g,' ').trim(),f++,i[f]={keys:new Set,urls:new Set,hasAnswer:!1,batches:[],prompt:p},void c.promptList.push(p);i[f]||(i[f]={keys:new Set,urls:new Set,hasAnswer:!1,batches:[],prompt:''});const u=i[f];if('assistant'===d&&o.content&&'thoughts'===o.content.content_type){const e=(o.content.thoughts||[]).map(e=>[e.summary,e.content].filter(Boolean).join(': ')).filter(Boolean).join(' | ');return void(e&&(y=(y?y+' | ':'')+e))}if('assistant'===d&&l.is_thinking_preamble_message&&o.content&&Array.isArray(o.content.parts)){const e=o.content.parts.filter(e=>'string'==typeof e).join(' ').replace(/\s+/g,' ').trim();return void(e&&(y=(y?y+' | ':'')+e))}if(('web.run'===o.recipient||'web'===o.recipient)&&o.content){const n='string'==typeof o.content.text?o.content.text:(o.content.parts||[]).filter(e=>'string'==typeof e).join('\n');g++,w={turn:f,pos:u.batches.length,entries:[],seen:new Map,note:y.slice(0,700)},y='',r[g]=w,u.batches.push(g);const i=s.length;if(n&&n.split(/\r?\n/).forEach(n=>((n,a,o,r)=>{if(!(n=n.trim()))return;const i=n.split('|'),c=i[0].trim();if(t.includes(c))return;if(!e.includes(c))return;let d='',h='',l='',p='';if('business'===c?(p=(i[1]||'').trim(),d=i.slice(2).filter(e=>''!==e).join(' ; ').trim()):(d=(i[1]||'').trim(),i.length>=3&&/^\d+$/.test(i[2].trim())?(h=i[2].trim(),l=(i[3]||'').trim()):i.length>=3&&(d=i.slice(1).join('|').trim())),!d)return;const u=/reddit/i.test(d+' '+l)?'yes':'no';s.push({n:s.length+1,batch:a,turn:r,type:c,query:d,days:h,domain:l,reddit:u,location:p,prompt:o,results:'',cited:'',locked:'',batchResults:'',batchCited:'',batchDomains:'',sources:[]})})(n,g,p,f)),s.length===i){const e=k(C,o.id,a),t=e?null:k(x,o.id,a),n=e||t;n?(e?c.savedRounds++:c.liveRounds++,n.q.forEach((t,a)=>s.push({n:s.length+1,batch:g,turn:f,type:n.t&&n.t[a]||'web',query:t,days:'',domain:'',reddit:/reddit/i.test(t)?'yes':'no',location:'',prompt:p,results:'',cited:'',locked:'',batchResults:'',batchCited:'',batchDomains:'',sources:[],qsrc:e?'saved':'live'}))):(c.hiddenQueries++,s.push({n:s.length+1,batch:g,turn:f,type:'web',query:0===w.pos?'query not captured: the bookmark was not running when this prompt was sent':'follow-up batch: ChatGPT does not send these queries',hidden:!0,days:'',domain:'',reddit:'no',location:'',prompt:p,results:'',cited:'',locked:'',batchResults:'',batchCited:'',batchDomains:'',sources:[],qsrc:''}))}return}'tool'===d&&w&&l.search_result_groups?(l.search_result_groups||[]).forEach(e=>(e.entries||[]).forEach(e=>{e&&e.ref_id&&null!=e.ref_id.turn_index&&(v[e.ref_id.turn_index]=w.pos),T(w,e,w.pos)})):'assistant'===d&&(l.search_result_groups&&(l.search_result_groups||[]).forEach(e=>(e.entries||[]).forEach(e=>{let t=e&&e.ref_id?e.ref_id.turn_index:null;null!=t&&null==u.batches[t]&&null!=v[t]&&(t=v[t]);const n=null!=t&&null!=u.batches[t]?u.batches[t]:u.batches[u.batches.length-1];null!=n&&T(r[n],e,r[n].pos)})),l.content_references&&((e,t,n,a)=>{const o=(e.metadata||{}).content_references||[],s=e=>{e&&t.urls.add(h(e))},r=e=>{e&&null!=e.turn_index&&t.keys.add(n+'|'+e.turn_index+'|'+e.ref_type+'|'+e.ref_index)};o.forEach(e=>{if(e&&e.type&&('grouped_webpages'===e.type||'webpage'===e.type||'webpage_extended'===e.type||'sources_footnote'===e.type||'news'===e.type||'products'===e.type)){let o;(e.items||[]).forEach(e=>{s(e.url),(e.refs||[]).forEach(r)}),(e.sources||[]).forEach(e=>{s(e.url),(e.refs||[]).forEach(r)}),e.url&&s(e.url);const i=/turn(\d+)([a-z]+)(\d+)/g,c=String(e.matched_text||'');for(;o=i.exec(c);){const e=a[o[1]];null!=e&&t.keys.add(n+'|'+e+'|'+o[2]+'|'+o[3])}}})})(o,u,f,v),(!0===o.end_turn||(l.content_references||[]).length)&&(u.hasAnswer=!0))}),c.prompts=f;const _=new Set,E=new Set;r.forEach(e=>{if(!e)return;const t=i[e.turn]||{keys:new Set,urls:new Set};e.entries.forEach(e=>{e.cited=t.keys.has(e.key)||t.urls.has(e.url),_.has(e.url)||(_.add(e.url),c.fetched++,l(e.host,'reddit.com')&&c.redditFetched++),e.cited&&!E.has(e.url)&&(E.add(e.url),c.cited++,l(e.host,'reddit.com')&&c.redditCited++)})});s.forEach(e=>{const t=r[e.batch]||{entries:[],turn:0},a=t.entries,s=!(!i[t.turn]||!i[t.turn].hasAnswer);if(e.known=s,e.age=null==e.age?'':e.age,e.citedAge='',e.note=t.note||'',e.batchResults=a.length,e.batchCited=s?a.filter(e=>e.cited).length:'',e.batchDomains=(e=>{const t={};return e.forEach(e=>{t[e.host]=(t[e.host]||0)+1}),Object.keys(t).sort((e,n)=>t[n]-t[e]).slice(0,3).map(e=>e+' '+t[e]).join(', ')})(a),o.includes(e.type))return;const c=(e=>{if(e.domain)return e.domain.replace(/^https?:\/\//,'').replace(/^www\./,'').split('/')[0].toLowerCase();const t=e.query.match(/site:([^\s"']+)/i);return t?t[1].replace(/^https?:\/\//,'').replace(/^www\./,'').split('/')[0].toLowerCase():''})(e);e.locked=c;const d=c?a.filter(e=>l(e.host,c)):n.includes(e.type)?a:null;if(!d)return;e.sources=d.slice().sort((e,t)=>t.cited-e.cited||e.host.localeCompare(t.host)||e.url.localeCompare(t.url)),e.results=d.length,e.cited=s?d.filter(e=>e.cited).length:'';const h=d.filter(e=>e.date).map(e=>pe(e.date));e.age=h.length?me(h):'';const p=s?d.filter(e=>e.cited&&e.date).map(e=>pe(e.date)):[];e.citedAge=p.length?me(p):'',e.note=t.note||''}),c.unknownTurns=i.filter(e=>e&&!e.hasAnswer).length},w=['n','batch','type','query','days','age','citedAge','domain','results','cited'],y={age:'fetched age',citedAge:'cited age'},b=['n','batch','days','age','citedAge','results','cited'],x=['days','domain'];let C='n',k=1;const v=(e,t)=>'query'===t?(e=>'business'===e.type?e.location+' : '+e.query:e.query)(e):'age'===t||'citedAge'===t?''===e[t]||null==e[t]?'':ue(e[t]):e[t],T=()=>{const e=s.slice();return e.sort((e,t)=>{const n=v(e,C),a=v(t,C);if(''===n&&''!==a)return 1;if(''===a&&''!==n)return-1;return(b.includes(C)?Number(n)-Number(a):String(n).localeCompare(String(a),void 0,{sensitivity:'base'}))*k||e.n-t.n}),e},_=e=>'"'+String(null==e?'':e).replace(/"/g,'""')+'"',E=(e,t)=>e.sources.filter(e=>!t||e.cited).map(e=>e.raw).join(' | '),S=e=>'live'===e.qsrc?'captured live':'saved'===e.qsrc?'saved chat':e.hidden?'not exposed':'saved chat',q=()=>[['n','batch','type','query','query_source','freshness_days','domain','results','cited','dated_pages','median_fetched_age_days','median_cited_age_days','pages_under_30_days','batch_note','reddit','location','locked_host','batch_results','batch_cited','batch_top_domains','sources','cited_sources','prompt','conversation_id','captured_at'].join(',')].concat(T().map(e=>{const t=(e=>e.sources.filter(e=>e.date).map(e=>pe(e.date)))(e);return[e.n,e.batch,e.type,e.query,S(e),e.days,e.domain,e.results,e.cited,t.length,t.length?me(t):'',e.citedAge,t.filter(e=>e<=30).length,e.note||'',e.reddit,e.location,e.locked,e.batchResults,e.batchCited,e.batchDomains,E(e,!1),E(e,!0),e.prompt,c.id,c.at].map(_).join(',')})).join('\n'),R=(e,t)=>{const n=new Blob(['\ufeff'+e],{type:'text/csv;charset=utf-8'}),a=document.createElement('a');a.href=URL.createObjectURL(n),a.download=t,document.body.appendChild(a),a.click(),setTimeout(()=>{URL.revokeObjectURL(a.href),a.remove()},1e3)},A=()=>{return'chatgpt-fanout-'+(e=c.title,(String(e||'').toLowerCase().normalize('NFKD').replace(/[^a-z0-9]+/g,'-').replace(/^-+|-+$/g,'').slice(0,60)||c.id.slice(0,8))+'-')+c.at.slice(0,16).replace(/[:T]/g,'-');var e},L=document.getElementById('fo-export');L&&L.remove();const P=document.createElement('div');P.id='fo-export',P.style.cssText='position:fixed;top:12px;right:12px;z-index:2147483647;box-sizing:border-box;width:min(1040px,96vw);max-width:calc(100vw - 24px);max-height:88vh;overflow:auto;resize:both;min-width:320px;min-height:160px;overscroll-behavior:contain;background:rgb(17,17,17);color:rgb(235,235,235);font:13px/1.4 -apple-system,Segoe UI,Helvetica,Arial,sans-serif;border:1px solid rgb(70,70,70);border-radius:10px;padding:14px;box-shadow:0 8px 30px rgba(0,0,0,.5);text-align:left';const M=document.createElement('div');M.style.cssText='display:flex;gap:8px;align-items:center;flex-wrap:wrap;position:sticky;top:-14px;z-index:5;background:rgb(17,17,17);margin:-14px -14px 10px;padding:14px 14px 10px';const G=document.createElement('strong');G.textContent='ChatGPT Fanout Explorer',G.style.cssText='font-size:14px';const j=document.createElement('span');j.textContent='v'+a,j.title='Build date of the code you are running. The install page always has the newest one.',j.style.cssText='font-size:11px;color:rgb(150,150,150);margin-right:auto';const N=e=>{const t=document.createElement('button');return t.type='button',t.textContent=e,t.style.cssText='background:rgb(43,43,43);color:rgb(255,255,255);border:1px solid rgb(90,90,90);border-radius:6px;padding:6px 10px;cursor:pointer;font:inherit',t},O=(e,t)=>{const n=e.textContent;e.textContent=t,setTimeout(()=>{e.textContent=n},1500)},z=N('Copy queries only'),D=N('Copy table (TSV)'),F=N('Download CSV'),I=N('Download sources CSV'),H=N('Live: on'),W=N('Refresh'),B=N('Close'),$=document.createElement('div');$.style.cssText='display:flex;gap:6px;align-items:center;margin:0 0 10px;border-bottom:1px solid rgb(70,70,70)';const U=e=>{const t=document.createElement('button');return t.type='button',t.textContent=e,t.style.cssText='background:none;color:rgb(170,170,170);border:none;border-bottom:2px solid transparent;padding:6px 10px;cursor:pointer;font:inherit;font-weight:600',t},V=U('Table'),J=U('Index'),Q=U('Types'),Z=N('Expand all');Z.style.cssText+=';margin-left:auto;padding:3px 8px;font-size:12px';const X=document.createElement('div');X.style.cssText='margin:6px 0 4px;color:rgb(190,190,190)';const K=document.createElement('div');K.style.cssText='margin:0 0 6px;color:rgb(235,235,235)';const Y=document.createElement('div');Y.style.cssText='margin:0 0 4px;color:rgb(235,235,235);word-break:break-word';const ee=document.createElement('div');ee.style.cssText='margin:0 0 10px;color:rgb(235,235,235);word-break:break-word';const te=document.createElement('div'),ne=document.createElement('div');ne.hidden=!0;const ae=document.createElement('div');ae.hidden=!0,M.append(G,j,z,D,F,I,H,W,B),$.append(V,J,Q,Z),P.append(M,$,X,K,Y,ee,te,ne,ae),document.body.appendChild(P);const oe='fo-export:pos:v1',se=()=>{try{return JSON.parse(localStorage.getItem(oe)||'null')}catch(e){return null}},re=e=>{const t=P.getBoundingClientRect(),n=se()||{},a={left:Math.round(t.left),top:Math.round(t.top)};e?(a.width=Math.round(t.width),a.height=Math.round(t.height)):n.width&&(a.width=n.width,a.height=n.height);try{localStorage.setItem(oe,JSON.stringify(a))}catch(e){}},ie=()=>{(()=>{const e=se(),t=e&&e.width?Math.min(Math.max(320,e.width),Math.max(320,innerWidth-16)):Math.min(1040,innerWidth-24);P.style.right='auto',P.style.width=t+'px',e&&e.height?(P.style.height=Math.min(Math.max(160,e.height),Math.max(160,innerHeight-16))+'px',P.style.maxHeight='none'):(P.style.height='',P.style.maxHeight=Math.max(220,innerHeight-24)+'px');const n=e?e.left:innerWidth-t-12,a=e?e.top:12;P.style.left=Math.min(Math.max(0,n),Math.max(0,innerWidth-120))+'px',P.style.top=Math.min(Math.max(0,a),Math.max(0,innerHeight-40))+'px'})();let e=P.getBoundingClientRect();const t=()=>(e=P.getBoundingClientRect(),e.top<-1||e.left<-1||e.right>innerWidth+1);t()&&P.parentNode!==document.documentElement&&document.documentElement.appendChild(P),t()&&(P.style.transform='none',P.style.top=parseFloat(P.style.top)-e.top+12+'px',P.style.left=parseFloat(P.style.left)-e.left+Math.max(12,innerWidth-P.offsetWidth-12)+'px')};ie(),addEventListener('resize',ie),M.style.cursor='move',M.title='Drag to move the panel. Double-click to put it back at the top right.',M.addEventListener('pointerdown',e=>{if(e.target.closest('button'))return;const t=P.getBoundingClientRect(),n=e.clientX,a=e.clientY,o=e=>{P.style.right='auto',P.style.left=Math.min(Math.max(0,t.left+e.clientX-n),Math.max(0,innerWidth-120))+'px',P.style.top=Math.min(Math.max(0,t.top+e.clientY-a),Math.max(0,innerHeight-40))+'px'},s=()=>{removeEventListener('pointermove',o),removeEventListener('pointerup',s),re()};addEventListener('pointermove',o),addEventListener('pointerup',s),e.preventDefault()}),M.addEventListener('dblclick',e=>{e.target.closest('button')||((()=>{try{localStorage.removeItem(oe)}catch(e){}})(),P.style.height='',ie())});let ce='';P.addEventListener('pointerup',()=>setTimeout(()=>{const e=P.offsetWidth+'x'+P.offsetHeight;ce&&e!==ce&&re(!0),ce=e},0)),ce=P.offsetWidth+'x'+P.offsetHeight;const de=e=>{const t='rgb(255,255,255)';[[V,'table'],[J,'index'],[Q,'types']].forEach(([n,a])=>{n.style.color=e===a?t:'rgb(170,170,170)',n.style.borderBottomColor=e===a?t:'transparent'});const n='table'===e;te.hidden=!n,X.hidden=!n,K.hidden=!n,Y.hidden=!n,ee.hidden=!n,Z.hidden=!n,ne.hidden='index'!==e,ae.hidden='types'!==e};V.onclick=()=>de('table'),J.onclick=()=>de('index'),Q.onclick=()=>de('types'),z.onclick=async()=>{try{await navigator.clipboard.writeText(T().map(e=>e.query).join('\n')),O(z,'Copied '+s.length)}catch(e){alert('Copy failed: '+e.message)}},D.onclick=async()=>{try{await navigator.clipboard.writeText(['n\tbatch\ttype\tquery\tfreshness_days\tdomain\tresults\tcited\treddit\tlocation\tprompt'].concat(T().map(e=>[e.n,e.batch,e.type,e.query,e.days,e.domain,e.results,e.cited,e.reddit,e.location,e.prompt].join('\t'))).join('\n')),O(D,'Copied')}catch(e){alert('Copy failed: '+e.message)}},F.onclick=()=>R(q(),A()+'.csv'),I.onclick=()=>R((()=>{const e=[['row_kind','n','batch','type','query','freshness_days','domain','results','cited','source_url','source_domain','source_cited','source_title','source_published','source_age_days','source_snippet','batch_note','prompt','conversation_id','captured_at'].join(',')];return T().forEach(t=>{e.push(['search',t.n,t.batch,t.type,t.query,t.days,t.domain,t.results,t.cited,'','','','','','','',t.note||'',t.prompt,c.id,c.at].map(_).join(',')),t.sources.forEach(n=>e.push(['source',t.n,t.batch,t.type,t.query,t.days,t.domain,t.results,t.cited,n.raw,n.host,t.known?n.cited?'yes':'no':'',n.title||'',n.date?ge(n.date):'',n.date?pe(n.date):'',n.snippet||'','',t.prompt,c.id,c.at].map(_).join(',')))}),e.join('\n')})(),A()+'-sources.csv');const he=(e,t,n,a,o)=>{const s=document.createElement('div');s.style.cssText='margin:0 0 10px;color:rgb(190,190,190)',s.textContent=t;const r=document.createElement('table');r.style.cssText='border-collapse:collapse;width:100'+String.fromCharCode(37);const i=document.createElement('tr');n.forEach(e=>{const t=document.createElement('th');t.textContent=e,t.style.cssText='text-align:left;border-bottom:1px solid rgb(70,70,70);padding:4px 6px;color:rgb(170,170,170);font-weight:600',i.appendChild(t)}),r.appendChild(i),a.forEach(e=>{const t=document.createElement('tr');e.forEach((n,a)=>{const s=document.createElement('td');s.textContent=n,s.style.cssText='border-bottom:1px solid rgb(42,42,42);padding:6px;vertical-align:top;'+(o[a]||'')+(0===a?';font-weight:600;color:rgb(255,255,255)':a===e.length-1?';color:rgb(190,190,190);font-style:italic;word-break:break-word':''),t.appendChild(s)}),r.appendChild(t)}),e.append(s,r)};he(ne,'Every line in the table is one search ChatGPT ran behind the scenes to answer this chat. Here is what each part means.',['Column','What it means','Example'],[['prompt (above the table)','The message you sent that started the searches. With several prompts in one chat, all of them are listed in order.',''],['n','The line number, in the order ChatGPT ran the searches. 1 is the first search of the chat.','n = 43 means it was the 43rd search in this chat.'],['batch','ChatGPT searches in batches. It sends a few searches together, reads what came back, then may send another batch. batch is the position of that batch in the answer: 1 for the first, 2 for the next.','All lines with batch = 2 were sent together, after ChatGPT had read the results of batch 1.'],['type','What kind of search it was. See the Types tab for the full list.','fast = a normal web search. business = a places search. image = a picture search. slow = a deeper web search.'],['query','The exact words ChatGPT sent to search, including any site: and quotes. This is the fan-out term. From September 2026 ChatGPT no longer saves it with the chat. It still streams the first batch of queries to the browser while it answers, so click the bookmark before you send a prompt and they are captured and kept for that chat. Work workspaces keep the queries in the saved chat. A batch with no query from either source says why in the query cell (the bookmark was not running when the prompt was sent, or it is a follow-up batch, which ChatGPT never streams) and still shows its pages and citations.','query = "Shrimp Shack Camden" reviews portion sauce birthday'],['days','How recent the pages had to be, in days, when ChatGPT still sent it. 30 = last month. 365 = last year. 3650 = last ten years. ChatGPT stopped sending this in September 2026, so the column only appears on chats from before then and stays hidden when every line is empty. For newer chats, fetched age and cited age show how old the pages actually were.','days = 30 on a chat from August 2026'],['domain','Only on chats from before September 2026, when ChatGPT still sent it, and hidden when every line is empty. When filled, ChatGPT only searched that one website. Empty = the whole web. A site: inside the query does the same job. Gone from new chats since September 2026, along with the query text, so this column is empty on them.','domain = reddit.com means only Reddit was searched. site:linkedin.com/jobs in the query means only LinkedIn jobs pages.'],['results','How many pages came back. For a line with a domain or a site:, it is the count from that website in that batch. For an open search, it is the count for the whole batch. ChatGPT records results per batch, not per query, so lines in the same batch that search the same place show the same number. Empty for business and image lines, whose results are not exposed.','results = 12 with domain = sexyfish.com means 12 pages from sexyfish.com came back. results = 0 with domain = reddit.com means the Reddit search returned nothing, so Reddit could not be cited from it.'],['cited','How many of those pages were shown as a source in the answer, either as a citation chip in the text or in the Sources list at the end. Empty while the answer is still being written, or when the answer for that prompt is not stored in the chat.','results = 11, cited = 3 means 11 pages came back and 3 were shown as sources. results = 84, cited = 0 means ChatGPT read 84 pages and credited none of them.'],['+ (first column)','Opens the line. First every page that was shown as a source in the answer, then the pages that were not, grouped by website: a website with several pages gets its own group with its full record in brackets, and the websites that returned a single page sit in one list. The age next to a page is its publication date read against today. Hover a page for its title and date. ChatGPT returns one pool of pages for the whole batch, so every query in a batch opens the same list.',''],['Show what ChatGPT read','A switch inside an opened line. On, each page shows the snippet ChatGPT was given for it, which is what it judged the page on before deciding whether to cite it, and the batch shows ChatGPT\'s own working note, the short thinking summary it wrote before running that batch, when the chat has one. Off by default so the list stays readable. Both are in the sources CSV as source_snippet and batch_note.',''],['Brands it searched (under the prompt)','The brands ChatGPT chose to check, read from the query text: a name in a query that matches one of the websites that came back, so words like SEO or AI never qualify and a brand it named but got no page from is not listed. The first batch of queries is written before a single result arrives, so these names come from what the model already holds: its training data, and your memory and custom instructions if they are on. A brand the model does not know cannot be in that first batch and can only enter through a generic query and whatever page answers it. Blocking the training crawlers on your own site makes that a little more likely, but most of what a model knows about a brand comes from other people\'s pages, so a well-discussed brand that blocks them is still named. Turn memory off when you measure, or your own history shapes the list.','Brands it searched: Semrush, Ahrefs, Sistrix, Sitebulb, Screaming Frog, seoClarity'],['headline (above the table)','Fetched = pages that came back across the whole chat, each counted once. Shown as sources = how many of those appeared in an answer. Reddit = the same two numbers for reddit.com only.','Fetched 204 pages, 19 shown as sources. Reddit: 57 fetched, 5 cited.'],['fetched age','How old the pages that came back are. ChatGPT records a publication date for many of the pages it fetches, roughly half in practice. Take those dated pages, work out each one\'s age from its date to today and sort them: the column shows the middle one, the median. 2mo means half the dated pages are two months old or newer. It is the median rather than the average so one ten-year-old page cannot drag the number. The small number after it is how many pages carry a date out of how many came back, so you can see what the median rests on. Hover the cell for the newest, the oldest and how many are from the last 30 days. Inside the line each dated page shows its own age. The days column was what ChatGPT asked for, fetched age is what it got. Reads undated when none of the pages carry a date.','fetched age = 2mo 26/58 means 26 of the 58 pages carry a date and half of those are two months old or newer. Hover: newest 6d, oldest 8.7y, 5 from the last 30 days'],['cited age','The same median, taken over the pages from this batch that were cited in the answer. Read it against fetched age: when the cited pages are newer than the pool they came from, freshness helped them get picked, and keeping your page current matters for this prompt. When the two are alike, it did not. Reads undated when the batch has cited pages but none of them carry a date, which is usual for product pages and help centres: The fetched pool had dates, the pages ChatGPT picked did not. The small number after it is how many of the cited pages carry a date, so 4mo 1/4 rests on a single page and is a weak read. Empty while the answer is being written or when nothing from the batch was cited.','fetched age = 2mo 26/58, cited age = 12d 3/3 means the pool was two months old in the middle and all three cited pages were from the last fortnight'],['highlighted rows','Lines where the query or the domain mentions Reddit.',''],['reddit (exports only)','yes if the query or the domain mentions Reddit, otherwise no.',''],['location (exports only)','For business lines, the place ChatGPT searched around.','location = West Finchley, London, UK'],['locked_host (CSV only)','The website a line was limited to, taken from the domain column or the site: in the query.','locked_host = linkedin.com'],['batch_results, batch_cited, batch_top_domains (CSV only)','For the whole batch the line belongs to: How many pages came back, how many were shown as sources, and the three websites with the most pages.','batch_top_domains = opentable.co.uk 4, wanderlog.com 1, tripadvisor.co.uk 1'],['sources, cited_sources (CSV only)','The pages behind the + for that line, and the ones that were cited, as lists separated by |.',''],['conversation_id, captured_at (exports only)','The chat ID from the URL, and when you exported.',''],['Copy queries only','Copies the query column, one per line, ready for a spreadsheet or a keyword tool.',''],['Copy table (TSV)','Copies the table plus the prompt as text you can paste straight into Google Sheets or Excel.',''],['Download CSV','One row per search line, with every column including the sources lists.',''],['Download sources CSV','The expanded view as a file: Each search line, followed by one row per page it got back, with source_url, source_domain and source_cited. Filter row_kind = source to work with the pages only.',''],['Sorting','Click a column header to sort by it, click again to reverse. Empty cells always go last. Copies and the CSV follow the sort you see. Sort by n to get back to the original order.',''],['Saved copies','Every chat the panel reads is kept in this browser. Reopening a chat shows the saved copy at once, without asking ChatGPT again. The newest 30 chats are kept. Refresh re-reads the live chat.','Status line: Saved copy from 06/09/2026 11:20. Refresh re-reads it.'],['429 / rate limiting','If ChatGPT refuses reads (error 429), the panel keeps showing what it has and waits before trying again, a minute at first, longer if it keeps happening. Live reads only happen while an answer is being written, so a finished chat costs one read, or none when a saved copy exists.',''],['Live: on / off','While on, the panel re-reads the chat every few seconds while ChatGPT is answering, and new lines appear as each batch of searching finishes. Once the answer is done it stops reading until you send another prompt. Open it on a new chat, send your prompt, and watch it fill. Turn it off to stop all reads. Refresh re-reads once.','Open a new chat, click the bookmark, type "best AI live chat tools for a small Shopify store", send, and the batches appear one by one.']],['width:170px','','width:34'+String.fromCharCode(37)]),he(ae,'Every kind of line ChatGPT writes to its search tool. "listed" types appear in the table, "ignored" ones are page actions or settings, not searches. "reported" means described by other researchers but not yet seen in these chats.',['Type','Kind','What it does','Shape','Example'],[['fast','search, listed','A normal web search. The workhorse, most lines are this type.','fast|query|days|domain, the last two optional','fast|Tidio pricing Lyro AI 2026 official|3650|tidio.com'],['slow','search, listed','A deeper web search. Seen so far on subreddit-specific queries, without a days window.','slow|query','slow|site:reddit.com/r/shopify Tidio Lyro Shopify review'],['business','search, listed','A places search, like a map search, for local questions. Results are not exposed in the payload, so results and cited stay empty.','business|location|categories separated by ;  or  business|location||names separated by ;','business|West Finchley, London, UK|seafood boil;Cajun seafood;lobster restaurant'],['image','search, listed','A picture search for the images shown in the answer. It does not feed the text. Results are not exposed.','image|query|days, days optional','image|Scott\u2019s Mayfair restaurant seafood'],['product','search, reported','A shopping search of product catalogues. Reported by other researchers, not yet seen in these chats. Listed if it appears.','product|query',''],['genui_run','other, reported','Builds a widget (a chart, a table) rather than searching. Reported by other researchers. Ignored.','',''],['open','action, ignored','Opens one of the results that came back, by its reference number, to read the page.','open|turnXsearchN','open|turn447052search12'],['find','action, ignored','Finds text inside an opened page.','find|turnXsearchN|text','find|turn447052search12|Badeparadies'],['click','action, ignored','Follows a link on an opened page.','click|turnXsearchN|link number','click|turn878257search1|0'],['length','setting, ignored','A hint for how long the answer should be. Always the last line of a batch.','length|short  or  length|medium','length|medium'],['Result types','for reference','Each page that comes back is tagged with where it came from: search (web), news, reddit (a separate Reddit source), business (places), image, view (a page ChatGPT opened). Reddit pages carrying their own tag is why Reddit can be fetched heavily and still cited rarely: it is a separate pool.','','']],['width:90px','width:110px','','width:200px;font-family:Menlo,Consolas,monospace;font-size:12px','font-family:Menlo,Consolas,monospace;font-size:12px']);const le=new Set,pe=e=>e?Math.max(0,Math.round((Date.now()/1e3-e)/86400)):null,ue=e=>null==e?'':e<1?'today':e<30?e+'d':e<365?Math.round(e/30)+'mo':Math.round(e/36.5)/10+'y',me=e=>{if(!e.length)return null;const t=e.slice().sort((e,t)=>e-t),n=t.length>>1;return 1&t.length?t[n]:Math.round((t[n-1]+t[n])/2)},ge=e=>{try{return new Date(1e3*e).toISOString().slice(0,10)}catch(e){return''}};let fe=!1;const we=e=>{const t=document.createElement('div');if(t.style.cssText='font-size:12px;line-height:1.35;color:rgb(200,200,200);padding:2px 0 8px 26px',o.includes(e.type))return t.textContent='Results for '+e.type+' lines are not exposed in the payload.',t;if(!e.sources.length)return t.textContent='Nothing came back for this search.',t;const n=e.sources.length,a=e.sources.filter(e=>e.cited).length,r=s.filter(t=>t.batch===e.batch).length,i=e.sources.filter(e=>e.date).map(e=>pe(e.date)),c=e.sources.filter(e=>e.cited&&e.date).map(e=>pe(e.date)),d=document.createElement('div');d.style.cssText='margin-bottom:6px;color:rgb(235,235,235)',d.textContent=n+' page'+(1===n?'':'s')+' came back for '+(r>1?'the '+r+' queries in this batch':'this batch')+(e.known?', '+a+' cited':'')+'.'+(i.length?' '+i.length+' carry a publication date: the middle one is '+ue(me(i))+' old, '+i.filter(e=>e<=30).length+' from the last 30 days'+(e.known&&c.length?', the middle cited one '+ue(me(c))+' ('+c.length+' of the '+a+' cited carry a date)':e.known&&a?', none of the '+a+' cited carry a date':'')+'.':'');const h=(e,t)=>{const n=document.createElement('div');n.style.cssText='white-space:nowrap;overflow:hidden;text-overflow:ellipsis;padding-left:'+t+'px';const a=document.createElement('span');a.textContent=e.cited?'\u2713 ':'\xb7 ',a.style.cssText=e.cited?'color:rgb(120,220,140);font-weight:700':'color:rgb(120,120,120)';const o=document.createElement('span');o.textContent=e.date?ue(pe(e.date)):'',o.title=e.date?'Published '+ge(e.date):'No date recorded for this page',o.style.cssText='display:inline-block;width:38px;color:rgb(185,185,185);font-size:11px';const s=document.createElement('a');if(s.href=e.raw,s.target='_blank',s.rel='noopener',s.textContent=(e=>{const t=String(e||'').replace(/^https?:\/\//,'').replace(/^www\./,'').split(/[?\x23]/)[0].replace(/\/$/,'');return t.length>90?t.slice(0,88)+'\u2026':t})(e.raw),s.title=(e.title?e.title+'\n':'')+(e.date?'Published '+ge(e.date)+'\n':'')+e.raw,s.style.cssText='color:'+(e.cited?'rgb(235,235,235)':'rgb(170,170,170)')+';text-decoration:none',n.append(a,o,s),fe){const a=document.createElement('div');a.style.cssText='padding-left:'+t+'px',a.appendChild(n),n.style.paddingLeft='0';const o=document.createElement('div');return o.style.cssText='color:rgb(150,150,150);font-size:11px;line-height:1.3;padding:0 0 4px 52px;white-space:normal;word-break:break-word',o.textContent=e.snippet?e.snippet:'No snippet recorded for this page.',a.appendChild(o),a}return n},l=e=>{const t=document.createElement('div');return t.textContent=e,t.style.cssText='color:rgb(235,235,235);font-weight:700;margin:6px 0 3px',t},p=(e,t)=>e.host.localeCompare(t.host)||e.url.localeCompare(t.url),u=document.createElement('div'),m=e.known?e.sources.filter(e=>e.cited).sort(p):[],g=e.known?e.sources.filter(e=>!e.cited):e.sources.slice();if(e.known){if(u.appendChild(l('Cited ('+m.length+')')),m.length){const e=document.createElement('div');m.length>6&&(e.style.cssText='column-count:2;column-gap:28px'),m.forEach(t=>e.appendChild(h(t,4))),u.appendChild(e)}else{const e=document.createElement('div');e.textContent='None of the pages from this batch were shown as a source.',e.style.cssText='color:rgb(170,170,170);padding-left:4px',u.appendChild(e)}u.appendChild(l(e.known?'Not cited ('+g.length+'), by website':'By website'))}const f={};e.sources.forEach(e=>{const t=f[e.host]||(f[e.host]={host:e.host,n:0,c:0,list:[]});t.n++,e.cited&&t.c++}),g.forEach(e=>f[e.host].list.push(e));const w={};Object.values(f).forEach(t=>{if(!t.list.length)return;const n=t.list.length+'|'+t.n+'|'+(e.known?t.c:0);(w[n]||(w[n]={shown:t.list.length,n:t.n,c:e.known?t.c:0,hosts:[]})).hosts.push(t)});const y=Object.values(w).sort((e,t)=>t.c-e.c||t.n-e.n),b=document.createElement('div');g.length>6&&(b.style.cssText='column-count:2;column-gap:28px'),y.forEach(t=>{const n=document.createElement('div');n.style.cssText='break-inside:avoid;margin:0 0 8px';const a=document.createElement('div');a.style.cssText='color:rgb(235,235,235);font-weight:600;white-space:nowrap;overflow:hidden;text-overflow:ellipsis';const o=t.hosts.sort((e,t)=>e.host.localeCompare(t.host)),s=t.shown+' page'+(1===t.shown?'':'s'),r=e.known&&t.c?' ('+t.n+' fetched, '+t.c+' cited)':'';a.textContent=1===o.length?s+': '+o[0].host+r:s+(1===t.shown?'':' each')+': '+o.length+' websites'+r,n.appendChild(a),1===o.length||1===t.shown?o.forEach(e=>e.list.slice().sort(p).forEach(e=>n.appendChild(h(e,4)))):o.forEach(e=>{const t=document.createElement('div');t.textContent=e.host,t.style.cssText='color:rgb(200,200,200);padding:2px 0 0 4px',n.appendChild(t),e.list.slice().sort(p).forEach(e=>n.appendChild(h(e,12)))}),b.appendChild(n)}),u.appendChild(b);const x=document.createElement('button');if(x.type='button',x.textContent=fe?'Hide what ChatGPT read':'Show what ChatGPT read',x.title='The snippet ChatGPT was given for each page, which is what it judged the page on, and its own working note for this batch when the chat has one',x.style.cssText='background:none;color:rgb(170,170,170);border:1px solid rgb(90,90,90);border-radius:5px;padding:2px 8px;margin-left:10px;cursor:pointer;font:inherit;font-size:11px',x.onclick=()=>{fe=!fe,ye()},d.appendChild(x),t.appendChild(d),fe){const n=document.createElement('div');n.style.cssText='margin:0 0 8px;padding:6px 8px;border-left:2px solid rgb(90,90,90);color:rgb(200,200,200);white-space:normal;word-break:break-word',n.textContent=e.note?'ChatGPT\u2019s note before this batch: '+e.note:'ChatGPT left no note before this batch.',t.appendChild(n)}return t.appendChild(u),t},ye=()=>{te.innerHTML='';const e=s.filter(e=>'yes'===e.reddit).length,t=s.length?s[s.length-1].batch:0,n=(c.liveRounds?' Queries for '+c.liveRounds+' batch(es) were captured live while ChatGPT answered and are kept for this chat in this browser.':'')+(c.savedRounds?' Queries for '+c.savedRounds+' batch(es) come from the saved chat.':'')+(c.hiddenQueries?' ChatGPT no longer saves the query text with the chat and only streams it while answering, so '+c.hiddenQueries+' batch(es) show pages and citations without the query, freshness window or site limit.'+(c.liveRounds?'':' Click the bookmark before you send a prompt and the queries are captured.'):'');if(X.textContent=s.length+' search line(s) in '+t+' batch(es) across '+c.prompts+' prompt(s). '+e+' mention Reddit.'+n+ze(),K.textContent=s.length?'Fetched '+c.fetched+' page(s), '+c.cited+' shown as sources in the answers. Reddit: '+c.redditFetched+' fetched, '+c.redditCited+' cited.'+(c.unknownTurns?' ('+c.unknownTurns+' prompt(s) have no finished answer yet, so their cited counts are left empty.)':''):'',Y.textContent=(()=>{const e=c.promptList||[];if(e.length<=1)return e.length?'Prompt: '+e[0]:'';const t=[...new Set(s.map(e=>e.turn))].sort((e,t)=>e-t),n=t.slice(0,4).map(t=>'Prompt '+t+': '+String(e[t-1]||'').slice(0,300)),a=t.length-n.length;return n.join('   ')+(a>0?'   \u2026 and '+a+' more prompt(s) with searches (all in the exports)':'')})(),ee.textContent=(()=>{const e=s.filter(e=>!e.hidden&&e.query).map(e=>e.query);if(!e.length)return'';const t=new Set;if((r.some(Boolean)?[].concat.apply([],r.filter(Boolean).map(e=>e.entries)):[].concat.apply([],s.map(e=>e.sources))).forEach(e=>{const n=e.host.split('.'),a=n.length>2&&/^(co|com|org|net|ac|gov)$/.test(n[n.length-2])?n[n.length-3]:n[n.length-2]||n[0];a&&t.add(a.toLowerCase())}),!t.size)return'';const n=[],a=new Set,o=e=>{if(e.length<3)return!1;for(const n of t){if(n===e)return!0;if(e.length>=6&&(n.startsWith(e)||n.endsWith(e))&&n.length-e.length<=3)return!0}return!1};return e.forEach(e=>{const t=e.replace(/["\u201c\u201d]/g,' ').split(/\s+/).filter(Boolean);for(let e=0;e<t.length;e++){if(!/^[A-Z][A-Za-z0-9.&+-]*$|^[a-z]+[A-Z][A-Za-z0-9]*$/.test(t[e]))continue;let s='';for(let n=Math.min(3,t.length-e);n>=1;n--){const a=t.slice(e,e+n);if(!a.every(e=>/^[A-Z][A-Za-z0-9.&+-]*$|^[a-z]+[A-Z][A-Za-z0-9]*$/.test(e)))continue;const r=a.join('').toLowerCase().replace(/[^a-z0-9]/g,'');if(o(r)){s=a.join(' '),e+=n-1;break}}s&&!a.has(s.toLowerCase())&&(a.add(s.toLowerCase()),n.push(s))}}),n.length?'Brands it searched: '+n.join(', '):''})(),Z.textContent=le.size?'Collapse all':'Expand all',!s.length)return;const a=document.createElement('table');a.style.cssText='border-collapse:collapse;width:100'+String.fromCharCode(37)+';table-layout:fixed';const o={n:40,batch:52,type:62,days:52,age:96,citedAge:88,domain:140,results:62,cited:52},i=w.filter(e=>!x.includes(e)||s.some(t=>''!==t[e]&&null!=t[e])),d=document.createElement('tr'),h=document.createElement('th');h.style.cssText='border-bottom:1px solid rgb(70,70,70);width:24px',d.appendChild(h),i.forEach(e=>{const t=document.createElement('th');t.textContent=(y[e]||e)+(C===e?1===k?' \u25b2':' \u25bc':''),t.title='age'===e?'Median age of the dated pages that came back for this batch. Hover a cell for the spread. Click to sort.':'citedAge'===e?'Median age of the dated pages from this batch that were cited in the answer. Click to sort.':'Click to sort by '+(y[e]||e),t.style.cssText='text-align:left;border-bottom:1px solid rgb(70,70,70);padding:4px 6px;color:'+(C===e?'rgb(255,255,255)':'rgb(170,170,170)')+';font-weight:600;white-space:nowrap;overflow:hidden;cursor:pointer;user-select:none'+(o[e]?';width:'+o[e]+'px':''),t.onclick=()=>{C===e?k=-k:(C=e,k=1),ye()},d.appendChild(t)}),a.appendChild(d),T().forEach(e=>{const t=le.has(e.n),n=document.createElement('tr');'yes'===e.reddit&&(n.style.background='rgb(70,45,10)');const o=document.createElement('td');o.style.cssText='border-bottom:1px solid rgb(42,42,42);padding:2px 0 2px 4px;vertical-align:top';const s=document.createElement('button');if(s.type='button',s.textContent=t?'\u2212':'+',s.title=t?'Hide the pages':'Show the pages that came back',s.style.cssText='width:18px;height:18px;line-height:16px;padding:0;background:rgb(43,43,43);color:rgb(235,235,235);border:1px solid rgb(90,90,90);border-radius:4px;cursor:pointer;font:12px/16px inherit',s.onclick=()=>{le.has(e.n)?le.delete(e.n):le.add(e.n),ye()},o.appendChild(s),n.appendChild(o),i.forEach(t=>{const a=document.createElement('td');a.textContent=v(e,t),a.style.cssText='border-bottom:1px solid rgb(42,42,42);padding:4px 6px;vertical-align:top;'+('query'===t?'word-break:break-word':'white-space:nowrap;overflow:hidden;text-overflow:ellipsis')+('cited'===t&&''!==e.cited&&e.cited>0?';color:rgb(120,220,140);font-weight:600':''),'domain'===t&&e.domain&&(a.title=e.domain),'query'===t&&(a.title='live'===e.qsrc?'Captured live while ChatGPT answered, kept for this chat in this browser':'saved'===e.qsrc?'From the saved chat':e.hidden?'ChatGPT did not send this query to the browser':'',e.hidden&&(a.style.cssText+=';color:rgb(140,140,140);font-style:italic'));const o=(e,t,n)=>{a.textContent='',a.appendChild(document.createTextNode(e+' '));const o=document.createElement('span');o.textContent=t+'/'+n,o.style.cssText='color:rgb(130,130,130);font-size:11px',a.appendChild(o)},s=()=>{a.textContent='undated',a.style.cssText+=';color:rgb(130,130,130);font-style:italic'};if('age'===t&&e.sources.length){const t=e.sources.filter(e=>e.date).map(e=>pe(e.date));t.length?(o(ue(me(t)),t.length,e.sources.length),a.title=t.length+' of the '+e.sources.length+' pages that came back carry a date. Median '+ue(me(t))+', newest '+ue(Math.min.apply(null,t))+', oldest '+ue(Math.max.apply(null,t))+', '+t.filter(e=>e<=30).length+' from the last 30 days.'):(s(),a.title='None of the '+e.sources.length+' pages that came back carry a publication date')}if('citedAge'===t&&e.sources.length){const t=e.sources.filter(e=>e.cited&&e.date).map(e=>pe(e.date)),n=e.sources.filter(e=>e.cited).length;a.title=e.known?n?t.length?t.length+' of the '+n+' cited pages carry a date. Median '+ue(me(t))+', newest '+ue(Math.min.apply(null,t))+', oldest '+ue(Math.max.apply(null,t))+'.':'None of the '+n+' cited pages carry a publication date, which is usual for product and help centre pages':'Nothing from this batch was cited':'The answer is not finished yet',e.known&&n&&(t.length?o(ue(me(t)),t.length,n):s())}n.appendChild(a)}),a.appendChild(n),t){const t=document.createElement('tr');'yes'===e.reddit&&(t.style.background='rgb(50,34,10)');const n=document.createElement('td');n.colSpan=i.length+1,n.style.cssText='border-bottom:1px solid rgb(42,42,42);padding:0 6px',n.appendChild(we(e)),t.appendChild(n),a.appendChild(t)}}),te.appendChild(a)};Z.onclick=()=>{le.size?le.clear():s.forEach(e=>le.add(e.n)),ye()};const be='fo-export:v1:',xe=e=>{const t=Object.keys(localStorage).filter(e=>e.startsWith(be));t.length<=e||t.map(e=>{let t='';try{t=(JSON.parse(localStorage.getItem(e))||{}).at||''}catch(e){}return{k:e,at:t}}).sort((e,t)=>e.at<t.at?-1:1).slice(0,t.length-e).forEach(e=>localStorage.removeItem(e.k))},Ce=e=>{try{const t=new Date(e);return t.toLocaleDateString()+' '+t.toLocaleTimeString().replace(/:\d\d(\s|$)/,'$1')}catch(t){return e}};let ke=null,ve='',Te=!1,_e='',Ee=null,Se=!0,qe=!1,Re=!1,Ae=!1,Le=0,Pe=0,Me=0,Ge=!1,je=!1;const Ne=()=>!!document.querySelector('button[data-testid="stop-button"]'),Oe=async()=>{if(ve)return ve;const e=await(await fetch('/api/auth/session',{credentials:'include'})).json();if(ve=e&&e.accessToken,!ve)throw new Error('no access token. Are you logged in?');return ve},ze=()=>Ge?' Saved copy from '+Ce(c.at)+'.'+(Re?' Refresh re-reads it.':' Reading the rest...'):Se?Re?' Live: waiting for a new prompt.':' Live, updated '+(new Date).toLocaleTimeString()+'.':' Live is off.',De=async e=>{if(!Te&&ke){Te=!0;try{const t=await fetch('/backend-api/conversation/'+ke,{headers:{Authorization:'Bearer '+await Oe()}});if(429===t.status){const e=1e3*Number(t.headers.get('Retry-After'));throw Me=e>0?e:Math.min(Me?2*Me:6e4,6e5),Pe=Date.now()+Me,new Error('429')}if(401===t.status||403===t.status)throw ve='',new Error('HTTP '+t.status+', retrying');if(!t.ok)throw new Error('HTTP '+t.status);Me=0,Pe=0;const n=await t.text(),o=n.length+'|'+(n.match(/"current_node":\s*"([^"]+)"/)||['',''])[1];o!==_e||e?(_e=o,Ge=!1,c.id=ke,c.at=(new Date).toISOString(),f(JSON.parse(n)),Re=!Ne()&&Date.now()>Le,ye(),(()=>{if(!c.id||!s.length)return;const e={at:c.at,build:a,meta:{title:c.title,prompts:c.prompts,promptList:c.promptList,fetched:c.fetched,cited:c.cited,redditFetched:c.redditFetched,redditCited:c.redditCited,unknownTurns:c.unknownTurns},rows:s.map(e=>({n:e.n,batch:e.batch,turn:e.turn,type:e.type,query:e.query,days:e.days,domain:e.domain,reddit:e.reddit,location:e.location,prompt:e.prompt,results:e.results,cited:e.cited,locked:e.locked,batchResults:e.batchResults,batchCited:e.batchCited,batchDomains:e.batchDomains,known:e.known,hidden:!!e.hidden,qsrc:e.qsrc||'',age:e.age,citedAge:e.citedAge,note:e.note||'',sources:e.sources.map(e=>[e.raw,e.host,e.cited?1:0,e.date||0,e.title||'',e.snippet||''])}))},t=()=>localStorage.setItem(be+c.id,JSON.stringify(e));try{t()}catch(e){try{xe(10),t()}catch(e){}}try{xe(30)}catch(e){}})(),s.length||(X.textContent='No search lines yet. If the answer is finished and this stays empty, it never searched the web, or the format moved again.'+ze())):(Re=!Ne()&&Date.now()>Le,s.length&&(X.textContent=X.textContent.replace(/ (Live|Saved copy).*$/,ze())))}catch(e){'429'===e.message?X.textContent='ChatGPT is rate limiting this browser (429). '+(s.length?'Showing the '+(Ge?'saved copy from '+Ce(c.at):'last read')+'. ':'')+'Next try in '+Math.ceil(Me/1e3)+'s.':X.textContent='Could not read the conversation: '+e.message}finally{Te=!1}}},Fe=async()=>{if(qe)return;const e=(()=>{const e=location.pathname.match(/\/c\/([0-9a-fA-F-]{16,})/);return e?e[1]:''})();e!==ke&&(ke=e,_e='',Re=!1,Le=0,Ge=!1,s.length=0,le.clear(),te.innerHTML='',K.textContent='',Y.textContent='',ke?(e=>{try{const t=JSON.parse(localStorage.getItem(be+e)||'null');return!(!t||!t.rows||(s.length=0,t.rows.forEach(e=>{e.sources=(e.sources||[]).map(e=>({raw:e[0],url:h(e[0]),host:e[1],cited:!!e[2],date:e[3]||0,title:e[4]||'',snippet:e[5]||''})),null==e.age&&(e.age=''),null==e.citedAge&&(e.citedAge=''),null==e.note&&(e.note=''),s.push(e)}),Object.assign(c,t.meta),c.id=e,c.at=t.at,je=t.build!==a,0))}catch(e){return!1}})(ke)?(Ge=!0,Re=!je&&!Ne(),ye()):X.textContent='Reading conversation '+ke+' ...':X.textContent=(window.__foTap?'Query capture is armed. ':'')+'Waiting for a chat. Send your prompt here and the searches, with their queries, will appear as ChatGPT runs them.');const t=Date.now(),n=Ne();n&&(Re=!1),Ae&&!n&&(Le=t+15e3),Ae=n;ke&&Se&&!Re&&t>=Pe&&'hidden'!==document.visibilityState?await De(!1):ke&&Se&&t<Pe&&!Te&&(X.textContent=X.textContent.replace(/Next try in \d+s\./,'Next try in '+Math.ceil((Pe-t)/1e3)+'s.')),Ee=setTimeout(Fe,n||t<Le?4e3:3e3)};H.onclick=()=>{Se=!Se,H.textContent=Se?'Live: on':'Live: off',Se&&(Re=!1),ye()},W.onclick=()=>{ke&&(Re=!1,Pe=0,De(!0))};const Ie=()=>{qe=!0,clearTimeout(Ee),removeEventListener('resize',ie),document.removeEventListener('keydown',He,!0),window.__foTap&&(window.__foTap.onCapture=null),P.remove()};window.__foTap&&(window.__foTap.onCapture=e=>{qe||ke&&e!==ke||(Re=!1,_e='')}),B.onclick=Ie;const He=e=>{if('Escape'!==e.key||!P.isConnected)return;const t=e.target,n=t&&t.tagName;'INPUT'===n||'TEXTAREA'===n||t&&t.isContentEditable||Ie()};document.addEventListener('keydown',He,!0),de('table'),Fe()})();

On GitHub at github.com/oritsimu-new/fanout-explorer: chatgpt-fanout-explorer.bookmarklet.txt is this standalone version as a text file, chatgpt-fanout-explorer.min.js is what the short version loads.

Version 2026-09-22.8, released 22 September 2026

The queries are back on new chats. ChatGPT stopped saving the query text with the chat in September 2026, but while it answers it still streams the first batch of queries to your browser. The bookmarklet now reads that stream as it passes and keeps the queries for that chat in your browser, so they stay when you reopen it. The one rule: Click the bookmark before you send the prompt. A chat you only open afterwards shows its pages and citations with the queries marked as not exposed, because ChatGPT never sent them, and the same goes for batches after the first, which ChatGPT does not stream. Work workspaces on paid accounts keep the queries in the saved chat and those are read too. Page age is new: the freshness window is gone, but ChatGPT records a publication date for many of the pages it fetches, so each page now shows how old it is and each batch shows the median age, how many pages are from the last 30 days and the median age of the cited ones. Opened lines lead with the cited pages, then group the rest by website, and a page age column gives each batch the median age of what came back. CSV file names now carry the chat title. New in 22.3: a cited age column next to fetched age, so you can see whether the cited pages are newer than the pool they came from. A "Show what ChatGPT read" switch inside an opened line reveals the snippet ChatGPT was given for each page and its own working note for the batch. A "Brands it searched" line under the prompt lists the brands ChatGPT chose to check, which come from what it already knows since the first batch is written before any result arrives. The days and domain columns only appear on chats that still have them. An age cell reads undated when the pages carry no publication date, which is usual for product and help centre pages, and a cited page now keeps its snippet when ChatGPT lists it twice. Lighter on the rate limit: the reader no longer keeps re-reading a chat that has a stopped or regenerated answer in its history, and it never reads from a background tab, so moving between chats no longer runs into ChatGPT's too many requests error. Nothing is sent anywhere: The capture happens inside the page and stays in your browser.

Which version are you running? Click your bookmark on a chat and look next to the title at the top of the panel. It shows the build date, like v2026-09-22.8. No date there means you are on a build older than 2026-09-22.8.

Getting this build. The short version fetches the newest code from GitHub each time you click it, so it picks this up on its own within the hour. The standalone version has all of its code inside the bookmark and never changes by itself: Drag or paste it again from this page to move to this build. Opening this page is not enough, the bookmark itself has to be replaced.

If clicking the bookmark does nothing

Right-click the bookmark, choose Edit and check the URL starts with javascript:. If it is empty, short, or the address of this page, the drag did not take: Use the copy button and paste the code as the bookmark URL.
On a company-managed browser, bookmarklets are often blocked by policy. Open chrome://policy or edge://policy and look for javascript://* under URLBlocklist. A personal browser profile gets around it.
Console route, if bookmarks are blocked but DevTools is not: On chatgpt.com, press F12 or Ctrl+Shift+I, open the Console tab, paste the standalone code and press Enter. Chrome asks you to type allow pasting the first time.
It has to be a chat page on chatgpt.com, logged in, in the browser where you installed it. The desktop apps have no bookmarks bar.

Use

Live: Open a new chat in Chrome, click the bookmark, then send your prompt. The searches appear a batch at a time while ChatGPT answers, and because the bookmark was already running it captures the queries as ChatGPT streams them. Once clicked, it keeps capturing every prompt you send in that tab, even with the panel closed.
Saved chat: Open the chat, click the bookmark. Chats you have read before show at once from a saved copy in your browser.
The prompt shows above the table. Click + on any line to see the pages it got back, ticks mark the cited ones. Expand all opens every line for a screenshot.
Click a column header to sort. Download CSV is one row per search, Download sources CSV adds one row per page underneath each search.
The Index tab explains every column, the Types tab every kind of search line. If ChatGPT rate limits you (429), the panel keeps what it has and retries later.

The other bookmarklets

There is one of these for each assistant plus a single page with all three: Fanout Explorer, all three. The code for all of them is on GitHub at github.com/oritsimu-new/fanout-explorer.

Sharing

Send a link to this page or the all-three page. It works in Chrome, Edge or Brave, on chatgpt.com (not the desktop app), logged in, on your own chats.

Built by

Orit Mutznik
https://www.oritmutznik.com

Follow or contact me for any queries or feedback
https://www.linkedin.com/in/oritsimu
https://www.x.com/oritsimu