ChatGPT SEO: the technical guide to ChatGPT search

How ChatGPT search finds and chooses sources#

ChatGPT “will choose to search the web based on what you ask,” per OpenAI’s search help article, and users can force a search with the Search tool. Semrush’s clickstream data puts search at 34.5% of prompts in February 2026. When it does search, ChatGPT “typically rewrites your query into one or more targeted queries” — OpenAI’s example turns “CCR8 for cancer” into “CCR8 immunotherapy drug development 2025.”

Those queries go to “third-party search providers, as well as content provided directly by our partners.” The current help article names Microsoft and Shopify. Independent reverse-engineering by Peec AI describes a blended stack underneath, including an OpenAI-built index it calls “Labrador” with vertical indexes for news, PDFs, video, local and shopping. OpenAI hasn’t confirmed the details — but its own crawler, OAI-SearchBot, exists to feed ChatGPT search, so plan for both Bing and OpenAI’s index.

1. 1

   ChatGPT decides whether to search

   Current, specific or local questions trigger a search; stable knowledge is answered from training. Location is inferred at country, state or city level and memories may shape the rewrite.

2. 2

   It fans the question out into targeted queries

   Until mid-2026 most prompts produced two or three sub-queries. Around 8 August 2026 that jumped: Nectiv measured 7.61 fan-outs per prompt, 64% using the site: operator to search specific brand, vendor and .gov domains directly.

   Your lever: Have an official page on your own domain for every fact a buyer checks: pricing, specs, policies, integrations, comparisons.

3. 3

   Providers and OpenAI's index return candidates

   Peec measured sources retrieved per prompt doubling from about 12 to about 24 after the August change. Ahrefs found roughly as many retrieved URLs go uncited as cited.

   Your lever: Be indexed by Bing and crawlable by OAI-SearchBot — both paths feed the pool.

4. 4

   Pages are read, not rendered

   Indexed pages come from OAI-SearchBot; live fetches for a conversation come from ChatGPT-User. OpenAI documents no JavaScript rendering, and Vercel’s crawler study saw OpenAI’s bot download scripts without running them.

   Your lever: Put the answer in the server-rendered HTML.

5. 5

   The answer cites the passages it used

   Citations appear as inline source pills and in a Sources panel. Since May 2026 ChatGPT also adds branded inline links — Profound saw responses with them rise from about 4.5% to over 20%, sending far more clicks to homepages.

   Your lever: Match your titles and opening sentences to the sub-query, and state the fact early.

OpenAI's crawlers and robots.txt#

OpenAI’s crawler documentation lists each bot with its own robots.txt token, and the settings are independent. The one that decides search visibility is OAI-SearchBot: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.”

A fourth bot, OAI-AdsBot, only visits landing pages submitted as ChatGPT ads, to check their safety and relevance. A robots.txt that stays in ChatGPT search while opting out of training:

# ChatGPT search — required to be cited
User-agent: OAI-SearchBot
Allow: /
 
# Model training — your call, independent of search
User-agent: GPTBot
Disallow: /

• Changes take about 24 hours to reach OpenAI’s search systems.
• When both bots are allowed, OpenAI “may use the results from just one crawl for both use cases” — so allowing both doesn’t double your crawl load.
• To keep a page out entirely, use noindex, not Disallow. A disallowed URL can still surface as a link and title found through providers, and the crawler must be allowed in to read the noindex.
• ChatGPT-User needs no rule to work, and a Disallow for it isn’t guaranteed to be honored. Block it at the WAF if you must.

ChatGPT’s agent — now ChatGPT Work’s cloud browser — sends a normal Chrome user agent but signs its requests with Web Bot Auth (RFC 9421), so you can verify it rather than block it:

Signature-Agent: "https://chatgpt.com"
Signature-Input: sig1=(...);keyid="...";tag="web-bot-auth"
Signature: sig1=:...:
 
# Public keys: https://chatgpt.com/.well-known/http-message-signatures-directory

Technical requirements#

OpenAI publishes one hard requirement — don’t block OAI-SearchBot — and very little else. The rest of this spec sheet comes from Microsoft’s documentation (Microsoft is a named search provider) and from large correlation studies, graded accordingly.

OAI-SearchBot allowedRequired

In robots.txt and at the CDN/WAF layer, including its published IP ranges. This is the only inclusion rule OpenAI documents.

Content in the initial HTMLRequired

OpenAI doesn’t document rendering, and the only direct test (Vercel/MERJ) found no JS execution. Treat client-rendered text as invisible.

Indexed by BingHelps

Verify in Bing Webmaster Tools and submit sitemaps. Its AI Performance report covers Copilot and Bing, not ChatGPT — but Bing’s index still feeds ChatGPT’s providers.

IndexNowHelps

Ping on every publish and update so Bing picks up changes in minutes. Microsoft recommends it for freshness.

SpeedHelps

SE Ranking found pages with first contentful paint under 0.4s averaged 6.7 citations against 2.1 for pages over 1.13s. Correlation, but consistent with live fetches timing out.

FreshnessHelps

Pages updated within three months averaged 6.0 citations against 3.6 (SE Ranking), and ChatGPT’s citations run 458 days newer than Google’s organic results (Ahrefs).

Product feed (ACP)Helps

For shopping answers. Products come from structured feeds via the Agentic Commerce Protocol and are ranked on availability, price, quality and whether you’re the maker or primary seller.

Sitemaps & canonicalsUnconfirmed

OpenAI doesn’t say whether OAI-SearchBot reads sitemaps or honors rel=canonical. Keep both correct for Bing and Google.

FAQ & other schemaNo effect

No OpenAI statement for general pages. SE Ranking found pages with FAQ schema averaged 3.6 citations against 4.2 without.

llms.txtNo effect

OpenAI has never said ChatGPT uses it; SE Ranking measured a “negligible” effect. Harmless, but not a priority.

To check what OpenAI’s crawler actually receives, request a page with its user agent and look for your key sentence in the raw HTML:

curl -s -A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot" \
  https://yoursite.com/pricing | grep -c "Plans start at"
 
# 0 = the text is rendered client-side, or your WAF is serving a challenge

What ChatGPT cites#

OpenAI says it “doesn’t set a fixed level of visibility for individual sites,” and there’s no inclusion form for web pages. What the large studies do show is consistent: ChatGPT rewards pages that match the sub-query, answer early, stay current, and belong to brands the wider web talks about.

• Match the sub-query

  Observed

  Cited pages’ titles matched fan-out queries at 0.656 cosine similarity, and natural-language URL slugs had an 89.8% citation rate against 81.1% without (Ahrefs, Apr 2026).

• Official pages on your own domain

  Observed

  Since August 2026, site: fan-outs target brand, vendor and .gov domains, and “official” is the second most common fan-out word.

• Answer in the first third

  Observed

  44.2% of ChatGPT citations came from the first 30% of a page, and definitional “X is…” sentences were about twice as likely to be cited (Growth Memo, 1.2M responses).

• Freshness

  Observed

  Cited URLs skew newer than search results, and recently updated pages earn more citations. Update substantively — never just the date.

• Brand mentions, especially YouTube

  Observed

  Across 75K brands, YouTube mentions correlated most with AI visibility (~0.74), branded web mentions next (0.66–0.71); domain rating was weak (Ahrefs, May 2026).

• Sections with substance

  Observed

  SE Ranking found sections under 50 words averaged 2.7 citations, against 4.6 at 120–180 words and 5.7 above 180. Depth between headings helps; one-line sections don’t.

Ads and shopping don't buy citations

ChatGPT has run ads since February 2026 — for Free and Go users in the US, then Canada, Australia, New Zealand, the UK, Mexico, Brazil, Japan and South Korea — and OpenAI says ads “do not influence the answers.” Shopping results are “not ads”: they come from structured product metadata, and since March 2026 merchants supply it through Agentic Commerce Protocol feeds while using their own checkout.

Tracking ChatGPT traffic#

OpenAI’s publisher FAQ confirms ChatGPT “automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs,” and the referrer is chatgpt.com. Since 13 May 2026, GA4’s default AI Assistant channel groups ChatGPT traffic automatically (medium ai-assistant). It isn’t retroactive, and some app traffic arrives with no referrer, so keep a custom channel as a backstop.

# Custom channel "AI – ChatGPT", placed above Referral
# Session source matches regex:
^(chatgpt\.com|chat\.openai\.com|openai\.com)$
 
# All major AI assistants
^(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com)$

Tag your own ChatGPT ads (utm_medium=cpc) and product feeds (utm_medium=feed) so paid and feed clicks don’t inflate your organic citation numbers. Then add the server-side view — ChatGPT-User hits are the closest thing to a live signal that a real prompt pulled your page:

# OpenAI bot hits by type
grep -oE "OAI-SearchBot|ChatGPT-User|GPTBot|OAI-AdsBot" access.log | sort | uniq -c
 
# Pages fetched live for ChatGPT conversations
grep "ChatGPT-User" access.log | awk '{print $7}' | sort | uniq -c | sort -rn | head -20

Verify IPs against searchbot.json, chatgpt-user.json and gptbot.json — user agents are easy to spoof. OpenAI offers no publisher analytics dashboard, so citations themselves still need a prompt panel: a fixed set of buyer questions, run weekly, recording who’s cited.

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

ChatGPT SEO: frequently asked questions#