First, how ChatGPT Search actually picks sources
ChatGPT Search launched on October 31, 2024 as a feature inside ChatGPT, built on a fine-tuned version of GPT-4o. When your question triggers a browse, it fetches live web results, reads a handful of pages, and returns an answer with inline numbered citations — typically three to six clickable sources. Those citations are the prize. Everything in this guide is about becoming one of them.
The key thing to understand: ChatGPT does not have its own web index. It leans on third-party search providers plus content from direct partners. OpenAI's help docs name Bing and Shopify as providers that may receive rewritten queries. OpenAI does not publish a ranking or weighting formula — so the practical playbook comes from where citations actually land, not from a published spec. That is also why ongoing AI citation tracking beats guessing.
The index is moving — do not bet on one engine
ChatGPT historically leaned on Bing, but it is shifting toward Google. Profound's analysis of 240M citations and 1,000 prompts found ChatGPT-to-Google index alignment rose from 12% (Apr 2025) to 33% (Jul 2025), while ChatGPT-to-Bing alignment fell from 26% to 8%. The honest caveat: most ChatGPT citations still match neither index exactly, so source selection is largely independent. The takeaway stands — you need to rank in both Bing and Google.
Want to know whether ChatGPT can already read and cite your site before you start? Run a free AI SEO audit — it checks crawlability and AI visibility in about 15 seconds.
The ChatGPT Search ranking checklist
Work these in order. The first one is the non-negotiable entry gate — get it wrong and nothing else you do will register.
- 1
1. Allow OAI-SearchBot (and whitelist it at the CDN)
OAI-SearchBot is the crawler that controls ChatGPT Search visibility — not GPTBot, not ChatGPT-User. Per OpenAI's bots docs, if you block OAI-SearchBot in robots.txt "your site will not be shown in ChatGPT search answers." Allow it in robots.txt and whitelist OpenAI's published crawler IP ranges at your CDN or WAF, since an aggressive AI-bot block is the most common silent killer of visibility. The three crawlers are independent: you can allow OAI-SearchBot for search while blocking GPTBot from training. Robots.txt changes take roughly 24 hours to register. Full breakdown in our guide to AI crawlers.
- 2
2. Rank in Bing first
This is the single highest-leverage lever. Seer Interactive analyzed 500+ citations across 100 queries and found 87%+ of ChatGPT Search citations matched Bing's top organic results, mostly positions 1–10. Submit to Bing Webmaster Tools and Bing Places, and treat Bing rankings as a primary KPI — not a legacy afterthought. If you are invisible in Bing's top 10, you are starting from behind in ChatGPT.
- 3
3. Rank in Google too — the index is drifting there
Google match was only 56% with a median rank of 17 in early data, but ChatGPT-to-Google index alignment is climbing fast (12%→33% in three months). Optimizing only for Bing is now a gap. Solid Google organic rankings are increasingly part of ChatGPT Search visibility, so do not let Google SEO slide while you chase Bing.
- 4
4. Lead with a direct, extractable answer
Open each page — and each section — with a direct 1–3 sentence answer, then expand below it. Use H2/H3 headings phrased as the exact questions users ask, plus short paragraphs, bullets, tables and an FAQ block. The model lifts self-contained, extractable chunks into its reply; content buried below a long intro rarely gets cited. Structure for extraction, not for scroll depth — that is the core of answer engine optimization.
- 5
5. Cite sources and add inline statistics
These are the validated GEO tactics. Write "According to Adobe's 2025 study, 55%…" rather than "55% of consumers…". Adding cited sources, named quotations and dated statistics are exactly the methods shown to lift visibility in the peer-reviewed GEO study — up to a 40% boost in generative-engine responses. See generative engine optimization for the full method. Owned numbers and named, sourced claims are what the model quotes verbatim.
- 6
6. Build off-site authority and brand mentions
ChatGPT favors content corroborated across trusted third-party domains — news sites, aggregators, directories and review platforms. Seer found the most-cited content types were affiliates, news sites and aggregators. Citation likelihood tracks your brand presence beyond your own site, not just on-page SEO, so digital PR and consistent listings are part of the work — and a big driver of share of voice.
- 7
7. Add schema.org structured data and rich metadata
Mark up your pages with Article, FAQPage, LocalBusiness and Product schema, and include rich metadata like hours, ratings and addresses. For local results, ChatGPT pulls from the public web and aggregators with visible ratings — it cannot read your Bing Places profile data directly, so put your NAP (name, address, phone) and hours in indexable HTML on the page, not only in a profile.
- 8
8. Keep content fresh and verifiable
Update high-value pages regularly and make sure every claim is independently confirmable. For local selection especially, "verifiable" and "linkable" are explicit filters — pages must be accessible without paywalls. A stale, unsourced, or gated page is a quiet disqualifier even when everything else is right.
- 9
9. Measure your ChatGPT citations on a schedule
You cannot improve what you cannot see, and ChatGPT's answers drift every time you ask. Track AI referral traffic in GA4 (ChatGPT referrals carry utm_source=chatgpt.com), and monitor your citations and share of voice over time. A single reading is noise; the trend line is the signal.
Steps 1, 2 and 9 are exactly what SourceWatch automates: it checks whether ChatGPT, Perplexity, Gemini and Claude can read and cite your site, then tracks your mentions and share of voice on a schedule — alongside the real AI-crawler and AI-referral traffic actually hitting your pages.
Track your ChatGPT visibility with SourceWatchWhy Bing still wins — and why you cannot ignore Google
This is the part most ChatGPT guides get wrong. They either tell you Bing is dead, or they ignore Bing entirely. Both are mistakes. Here is the real picture from the citation data.
87%+
Share of ChatGPT Search citations that matched Bing's top organic results — mostly positions 1–10 (Seer Interactive, 500+ citations across 100 queries, 2025)
In the same study, Google's match rate was only 56%, with a median Google rank of 17 — meaning Google pages that did get cited often ranked well below page one. So as of early 2025, Bing's top results were by far the strongest single predictor of being cited in ChatGPT. That is the lever to pull first.
But the index is moving toward Google
Profound's analysis of 240M citations and 1,000 prompts found ChatGPT-to-Google index alignment rose from 12% in April 2025 to 33% in July 2025, while ChatGPT-to-Bing alignment fell from 26% to 8% over the same window. The trend is unmistakable. The nuance worth keeping honest: most ChatGPT citations still match neither index exactly, so source selection is largely independent of either engine. The safe conclusion is simply to rank in both.
| Signal | What the data shows | Source |
|---|---|---|
| Bing top-10 match | 87%+ of ChatGPT Search citations | Seer Interactive (2025) |
| Google match | 56%, median rank 17 | Seer Interactive (2025) |
| ChatGPT↔Google index alignment | 12% → 33% (Apr→Jul 2025) | Profound (240M citations) |
| ChatGPT↔Bing index alignment | 26% → 8% (Apr→Jul 2025) | Profound (240M citations) |
| GEO tactics (cite, quote, stats) | Up to +40% visibility | GEO paper (arXiv) |
The practical rule
Rank in Bing first (highest-leverage today), Google second (rising fast). If your AI SEO program only touches one search engine, that is your gap — close it before optimizing anything else.
The three OpenAI crawlers (and why people block the wrong one)
OpenAI runs three separate crawlers, and conflating them is how good sites accidentally disappear from ChatGPT Search. Each is governed independently, so blocking one does not block the others.
- **OAI-SearchBot** — controls ChatGPT Search visibility. Block it and you will not appear in ChatGPT search answers. This is the one you must allow.
- **GPTBot** — used for model training. Blocking it keeps your content out of future training data but has no effect on whether you show up in search.
- **ChatGPT-User** — fetches a page live when a user action triggers it. It is not governed by robots.txt.
The most common mistake
Teams block GPTBot to opt out of training, then assume they are still searchable. They are — but if a blanket "AI bot" rule or a WAF also caught OAI-SearchBot, ChatGPT Search visibility is silently gone. You can absolutely allow OAI-SearchBot for visibility while blocking GPTBot from training. Check both lines independently, and remember robots.txt changes take about 24 hours to register.
Not sure which bots are reaching you? SourceWatch captures first-party AI-crawler traffic and flags verified vs spoofed hits, so you can confirm OAI-SearchBot is actually fetching your pages — see AI crawlers for how to read the logs.
Common mistakes that keep good sites out of ChatGPT
Most "we are not showing up in ChatGPT" cases are one of these. Check them before you write a single new word.
- **Blocking OAI-SearchBot** — usually via a blanket AI-bot block or an aggressive WAF rule. This silently kills all ChatGPT Search visibility.
- **Optimizing only for Google and ignoring Bing** — Bing remains the top citation predictor at 87%+. Bing is not legacy here.
- **Burying the answer below a long intro** — non-extractable content rarely gets cited. Lead with the direct answer.
- **Unsourced stats and vague claims** — "55% of consumers" loses to "According to Adobe's 2025 study, 55%…". Dated, cited figures win.
- **Assuming you can apply for a "publisher partnership"** — there is no open application. Those are direct licensing deals with large publishers (AP, Axel Springer, Le Monde). For everyone else the path is authority plus crawlability.
- **Treating llms.txt as a confirmed ranking factor** — it is an emerging convention for inference-time context, not an official OpenAI ranking signal, and not a robots.txt or sitemap replacement. Publish it if you like, but do not expect it to move ChatGPT.
- **Conflating crawlers** — blocking GPTBot to stop training does NOT remove search visibility. That is governed by OAI-SearchBot, which is independent.
For the broader cross-engine playbook, see the pillar on how to rank in ChatGPT. For the e-commerce surface, see how to rank in ChatGPT Shopping. To go beyond citations to recommendations, see ChatGPT SEO, and to keep score over time, how to track AI mentions.