SEO for ChatGPT: How to Earn Citations in ChatGPT Answers

Getting cited by ChatGPT comes down to three things. Let OpenAI’s crawlers in. Publish answers a machine can lift cleanly. Get talked about on the sites ChatGPT already trusts. The third one carries the most weight, and almost nobody budgets for it. One clarification before we go further: SEO for ChatGPT means optimizing your brand so ChatGPT cites it, not using ChatGPT to write your meta tags.

The mechanics changed hard across 2025 and 2026. ChatGPT no longer leans on Bing the way it did at launch. It runs its own index, its own page cache and its own crawler fleet. Most guidance written before mid-2026 describes a system that has already been replaced.

This guide covers what is true as of August 2026. How ChatGPT retrieves sources. What each of OpenAI’s four user agents actually does. Which signals decide citations. Why third-party mentions beat your own pages. And a checklist you can run this week.

SEO for ChatGPT vs. using ChatGPT for SEO

Two very different jobs share one search phrase. Let’s separate them in thirty seconds.

Using ChatGPT for SEO means prompting the chatbot to draft titles, cluster keywords or outline a brief. Useful. It is a productivity trick, not a channel strategy.

SEO for ChatGPT points the other way. You optimize your site, your entity footprint and your off-site reputation so ChatGPT retrieves and names you when a buyer asks a question in your category. Practitioners file this under answer engine optimization, GEO or LLM SEO. Same discipline, different labels.

Everything below is the second one. If you want a robot to write your blog posts, you are in the wrong place. If you want ChatGPT to say your brand name in front of somebody with a credit card, keep reading.

How ChatGPT actually sources answers in 2026

Short version: SEO for ChatGPT now means optimizing for a retrieval stack OpenAI owns end to end. Bing is no longer the story.

The Bing era, and what replaced it

OpenAI launched ChatGPT search on 31 October 2024 and opened it to all users on 5 February 2025. At the time it leaned on third-party providers, with Microsoft’s index doing the heavy lifting. OpenAI’s help documentation still says ChatGPT search “sometimes partners with other search providers” and rewrites your question into one or more targeted queries before sending them out.

The balance has shifted since. OpenAI now runs its own crawler, OAI-SearchBot, feeding its own web index. Research published in Search Engine Land on 17 August 2026 by the agency RESONEO analysed 1,200 ChatGPT answers, 88,000 search results and 26,900 distinct pages, using a browser extension plus canary pages on their own domain with full server logs. It found an OpenAI-owned index supplying roughly three quarters of citations in ChatGPT’s free instant mode. Only 1.5% of the URLs that index returned appeared in Bing’s top 20 for the same query.

So “just get indexed in Bing” is stale advice. Bing coverage is not worthless, but it is no longer the mechanism. Treat OAI-SearchBot access as the thing that actually gates you. This is the single biggest correction to make in any SEO for ChatGPT plan drafted before 2026.

Three retrieval layers: index, cache, live open

The same study describes three layers, and they behave very differently.

• Discovery index. Finds candidate URLs. In instant mode this is where nearly every citation originates. The model receives your full title tag untruncated, plus roughly the first 200 characters of body text after your H1. Meta descriptions are ignored entirely.
• Reading cache. A shared store of full pages converted to Markdown. Copies stay fresh for around 30 minutes on frequently requested URLs, then sit stale until a background refresh. Cache-Control and noindex headers get ignored at this layer.
• Live open. Rare, and mostly limited to Thinking mode. It matters more than its frequency suggests. In the RESONEO sample, 61,332 URLs surfaced, 5,032 became lead sources and only 759 pages were actually opened. Pages that got opened earned a citation 74% of the time. Pages that surfaced but were never read earned one 7% of the time.

Two consequences fall out of that. Your title and your first 200 characters do the qualifying work, so stop writing truncation-optimized labels and start writing self-contained sentences. And getting opened is the real conversion event, which is decided by how relevant and specific your snippet looks sitting in a candidate list next to forty others.

Practical version: move category labels, breadcrumbs, bylines and dates below your H1. Whatever sits immediately after that heading is the audition tape. Get that right and the rest of your SEO for ChatGPT work has something solid to stand on.

SEO for ChatGPT starts with OpenAI’s four user agents

Four bots, four jobs. Confusing them is the single most common technical error in SEO for ChatGPT, and it is the one that silently costs the most. Here is the current lineup, verified against OpenAI’s official crawler documentation.

User agent	What it does	Used for model training?	What blocking it costs you
OAI-SearchBot	Crawls pages so they can be surfaced in ChatGPT’s search features.	No	OpenAI states your site “will not be shown in ChatGPT search answers.” This is the one that matters.
GPTBot	Crawls content that may be used to train OpenAI’s generative foundation models.	Yes	Your content is excluded from training. Indirect effects on citations appear real (see below).
ChatGPT-User	Fetches a page on demand when a user, a Custom GPT or an agent action asks ChatGPT to open a link.	No	Little indexing impact. OpenAI notes robots.txt “may not apply” because the fetch is user-initiated.
OAI-AdsBot	Validates the safety of web pages submitted as ads on ChatGPT.	No	Your pages cannot be validated for ChatGPT ad placements.

The version tokens you will see in logs: OAI-SearchBot/1.4, GPTBot/1.4, ChatGPT-User/1.0 and OAI-AdsBot/1.0. Each ships a documentation URL in the string, so +https://openai.com/searchbot and friends are easy to grep for.

User-agent strings are trivial to spoof. OpenAI publishes IP ranges as JSON at openai.com/searchbot.json, /gptbot.json, /chatgpt-user.json and /adsbot.json. Verify against those before you conclude anything from a log line, and before you let a WAF rule ban a range you actually want.

Should you block GPTBot?

This is the most consequential single decision in SEO for ChatGPT, and the honest answer is that it depends on whether you are protecting a licensing position or chasing visibility. If you are chasing visibility, blocking looks expensive. A July 2026 analysis by cloro of 1,058 prominent domains found that sites blocking GPTBot showed a median citation propensity of 0.003 in ChatGPT, against 0.417 for sites allowing it. The authors call the finding directional rather than conclusive, and it reflects their own prompt mix. But the pattern held per-engine: blocking Perplexity’s bot depressed Perplexity citations and not ChatGPT’s, which is what you would expect from a causal relationship rather than a coincidence.

The same dataset showed how uneven adoption is. Roughly 13.9% of those domains blocked GPTBot, while only 3.4% blocked OAI-SearchBot. Plenty of teams blocked the training crawler on principle and left the retrieval crawler open. That is a defensible position. Blocking both by accident is not.

One timing note before you start testing: OpenAI says it can take about 24 hours for a robots.txt change to reach its search systems. Do not draw conclusions an hour after you edit the file.

What actually determines which sources ChatGPT cites

There is no published ranking algorithm. OpenAI says only that ChatGPT “ranks search results using multiple factors” and that “placement is not guaranteed.” What we do have is a lot of observed citation data, and it points somewhere consistent.

Ahrefs analysed ChatGPT’s 1,000 most-cited pages from September 2025. The profile of a cited page looked like this:

• Very high authority. 65.3% of cited pages sat on domains with a Domain Rating of 81 or above. The median was 90.
• Freshness matters more than you would guess. Around 76.4% had been updated within the previous month.
• Google rankings are not a prerequisite. 28.3% of cited pages had zero organic search visibility. Of the pages that did rank, 52.1% sat in the top three.
• Reference beats marketing. Wikipedia accounted for 29.7% of the top-cited pages, homepages and landing pages 23.8%, educational pages 19.4%, reviews 5.8%.

Read those four lines together and you have something close to a scoreboard for SEO for ChatGPT. Authority and recency are table stakes. Ranking on Google helps but is neither necessary nor sufficient. And the page types that win are reference-shaped, not sales-shaped.

Retrievability is a separate problem from quality

Plenty of excellent pages never get cited because the crawler cannot use them. Three hard technical limits are worth internalising for any serious SEO for ChatGPT effort.

• OpenAI’s fetcher does not execute JavaScript. Client-side rendered content is invisible. If your key answer arrives via hydration, it does not exist.
• Pages over 4 MB get rejected outright with an HTTP 400 and nothing is read at all. Heavy pages fail silently.
• The cache ignores noindex. A page you thought was hidden may still be sitting in ChatGPT’s reading cache as Markdown.

Fix retrievability before you touch content. Otherwise you are polishing something no machine will ever see. A quick GEO audit catches most of this in one pass.

Structure your answers to be liftable

The pattern that works is boring and consistent. Ask the question as a heading. Answer it in the next two sentences, in plain declarative language, with the entity named. Then support it. Do not bury the answer under three paragraphs of context. ChatGPT is extracting a claim, not appreciating your build-up.

Give it specifics it can quote: numbers, dates, named comparisons, prices, limits. Vague copy is unquotable copy. If a sentence could describe any of your competitors equally well, it will never be the sentence that gets cited.

Why third-party mentions beat your own site

This is the part most teams get wrong, and it is where the leverage sits. It is also the least automatable half of SEO for ChatGPT. ChatGPT cites the web talking about you far more often than it cites you talking about you.

According to Ahrefs’ most-cited-domains tracker, updated 21 July 2026, reddit.com held a 16.7% mention share of ChatGPT citations, ahead of en.wikipedia.org at 8.9%. Forbes came third at 3.3%. Consumer Reports, Healthline and Walmart all placed in the top ten. YouTube sat sixteenth at 1.8%.

Two brand-name publishers and a forum account for a quarter of everything ChatGPT cites. Your domain is competing for a sliver of what remains.

Ahrefs’ page-level study made the same point from another angle: only about 32.3% of the most-cited pages were of a type a marketer could realistically influence through outreach, such as educational content, reviews, news and blogs. The other two thirds were reference and organizational pages.

What to do with that

1. Get into the comparison and “best X” articles in your category. Not your own listicle. Somebody else’s. Find the pages ChatGPT already cites for your money prompts, then pitch the people who own them.
2. Earn genuine review-site presence. G2, Capterra, Trustpilot, Consumer Reports, whatever the equivalent is in your vertical. These pages get retrieved constantly because they are structured, dated and comparative.
3. Participate on Reddit honestly. Answer questions as a practitioner from a real account with history. Astroturfing gets removed, and removed threads cite nobody. The upside is real but it is a twelve-month game, not a campaign.
4. Fix your Wikipedia and Wikidata footprint if you legitimately qualify. Do not create a promotional article. Do make sure that where your entity already exists, the facts are correct and sourced.
5. Keep your named facts consistent everywhere. Same company name, same founding year, same product names, same pricing across your site, your review profiles, your press coverage and your schema. Contradictions make an entity harder to resolve, and unresolved entities get skipped.

A caution on Reddit specifically. Citation shares move fast and can move violently. In mid-August 2026 the GEO tracking vendor Promptwatch reported Reddit’s daily share of ChatGPT search citations falling from a 3.83% average across 18 July to 7 August down to roughly 0.52% between 14 and 17 August. Promptwatch flagged the numbers as provisional and OpenAI has said nothing publicly about it. Treat it as a reminder rather than a fact: any single platform can be reweighted overnight, so do not build your entire off-site strategy on one property.

Product and shopping surfaces: where ordinary SEO for ChatGPT stops

If you sell physical products, there is a second, separate pipeline, and normal page-level optimization will not get you into it. Products surface through structured feeds, not through crawled web pages.

OpenAI launched Instant Checkout and the open-source Agentic Commerce Protocol on 29 September 2025, starting with Etsy’s US sellers and Shopify merchants including Glossier, SKIMS, Spanx and Vuori. The stated rule at launch was that “product results are organic and unsponsored, ranked purely on relevance to the user,” and that Instant Checkout items are not given preference in results.

That model then changed. On 24 March 2026 OpenAI announced a pivot away from running checkout itself, saying the first version of Instant Checkout “did not offer the level of flexibility that we aspire to provide,” and that merchants could use their own checkout experiences while OpenAI concentrated on product discovery. Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair were named among the integrated retailers, and Walmart shipped a ChatGPT app with account linking and Walmart payments.

How merchants actually get in

Through the product feed, not the crawler. OpenAI’s commerce documentation describes a secure, regularly refreshed CSV or JSON feed carrying identifiers, descriptions, pricing, inventory, media and fulfilment options. You submit a sample feed for validation, then push daily snapshots. Optional attributes such as reviews and media are described as improving ranking, relevance and trust. Applications go through chatgpt.com/merchants, and access is still partner-gated.

Three things follow. Feed hygiene is now a ranking factor, so stale prices and wrong stock status cost you placements. Review volume and quality feed the comparison view, which is where the buying decision happens. And OpenAI is not the merchant of record, so you keep control of fulfilment, tax and payment, and you keep the customer relationship.

If you are ecommerce, run your feed work and your content work as two parallel tracks. They barely overlap.

The SEO for ChatGPT checklist

Here is the whole SEO for ChatGPT checklist. Run it in order. The first three items are gating, and skipping them makes everything after them pointless.

1. Allow OAI-SearchBot in robots.txt. Explicitly. Then confirm no WAF, CDN rule or bot-management product is blocking OpenAI’s published IP ranges. Wait 24 hours before you evaluate anything.
2. Decide GPTBot deliberately. Block it if you have a licensing position to protect. Otherwise allow it, given the observed relationship between blocking and citation loss.
3. Confirm your key content is in the server-rendered HTML. Fetch your page with JavaScript disabled. If the answer is not there, the crawler never sees it. Also check no important template exceeds 4 MB.
4. Rewrite title tags as self-contained sentences. The full title reaches the model untruncated, so a title that reads as a complete claim beats one padded with pipes and brand suffixes.
5. Rebuild the first 200 characters after each H1. Direct answer, entity named, no throat-clearing. Push dates, breadcrumbs and category chips below it.
6. Convert your best pages into question-shaped sections. H2 asks, first two sentences answer, rest supports. Add a real FAQ block with distinct questions people actually search.
7. Ship Organization, Product, Article and FAQPage schema. Structured data will not force a citation, but it makes entity resolution cheaper and it is close to free. Use a schema generator if you are doing it by hand.
8. Update your highest-value pages on a schedule. Given how strong the recency skew is, a genuine quarterly refresh with new dates, numbers and examples is worth more than three new thin posts.
9. Publish original data nobody else has. A survey, a benchmark, a pricing teardown. Original numbers are the most citable asset on the internet because there is no substitute source.
10. Build the off-site layer. Reviews, third-party comparisons, credible Reddit participation, correct Wikidata. This is the slow, expensive, high-leverage part.
11. Consider llms.txt, but keep expectations low. No AI provider has publicly confirmed using it for retrieval, and no correlation with citations has been demonstrated. It costs an afternoon and it does not hurt. See our take on what llms.txt is and is not before you spend a sprint on it.
12. Track prompts, not just pages. Pick the 20 to 50 questions your buyers actually ask and monitor who gets named. A ChatGPT rank tracker automates the repetitive part.

Sequence matters more than volume here. Most SEO for ChatGPT programmes stall because they start at step 9 while step 1 is quietly broken.

Five mistakes that quietly kill your citations

Nearly every SEO for ChatGPT audit turns up at least two of these. Each one is invisible until you go looking for it.

1. Blocking every AI bot with one rule. A blanket Disallow for anything matching “GPT” or “AI” catches OAI-SearchBot too. That is a full exit from ChatGPT search, made by accident.
2. Assuming Bing rankings equal ChatGPT visibility. They correlated once. The overlap between OpenAI’s own index results and Bing’s top 20 is now small enough that planning around Bing is planning around the wrong system.
3. Treating llms.txt as the strategy. It has become the cargo cult of this discipline. As one commenter in the SEO community put it, if nobody mentions you anywhere, schema and llms.txt probably are not your first problem.
4. Writing for word count. Long preambles push your answer past the window the discovery layer actually reads. Front-load or lose.
5. Measuring referral traffic only. Most ChatGPT citations never produce a click. If your only KPI is sessions from chatgpt.com, you will conclude this channel does not work while your competitor gets named in every answer.

How long SEO for ChatGPT takes, and how to know it works

Expect nothing for the first few weeks, then movement on retrievability fixes within roughly a month, and movement on off-site reputation over two to three quarters. Technical unblocking is fast. Entity building is not.

Set the measurement up before you start, or you will not be able to prove anything. The short version: pick a fixed prompt set, run it on a schedule from a clean session, and log whether you are cited, mentioned without a link, or absent. Then watch server logs for OAI-SearchBot and ChatGPT-User hits as a leading indicator, since crawl usually precedes citation.

We covered the full measurement workflow separately, including how to separate genuine citations from hallucinated ones, in how to check if ChatGPT cites your site. Read that alongside this one. This article is about earning the citation; that one is about proving you got it.

One last framing. SEO for ChatGPT is not a new channel bolted onto search. It is the same trust problem you have always had, evaluated by a system that reads faster than a human, forgets nothing, and mostly listens to what other people say about you.

Frequently asked questions

Does ChatGPT still use Bing to search the web?

Only partly. OpenAI’s own documentation says ChatGPT search “sometimes partners with other search providers,” and Microsoft is named as one. But research published in August 2026 found OpenAI’s own index supplying about three quarters of citations in free instant mode, with just 1.5% of those URLs appearing in Bing’s top 20. Optimizing purely for Bing is no longer a reliable route in.

Should I block GPTBot but allow OAI-SearchBot?

That is a legitimate configuration and roughly 14% of prominent domains do something like it. It keeps your content out of model training while preserving eligibility for ChatGPT search results. Be aware that observed data from July 2026 shows domains blocking GPTBot getting cited far less often overall, so it is a trade-off rather than a free lunch.

How long does it take for ChatGPT to cite a new page?

There is no fixed crawl-to-citation timeline. Robots.txt changes take about 24 hours to reach OpenAI’s search systems, and cached copies of popular pages refresh roughly every 30 minutes. In practice, technically retrievable pages on established domains can appear within days, while a new domain with no third-party mentions can wait months.

Does llms.txt help you get cited by ChatGPT?

There is no evidence that it does. No major AI provider has publicly confirmed using llms.txt for retrieval, and Google has said equivalent files neither help nor harm visibility. It takes almost no effort and causes no damage, so treat it as optional hygiene rather than a lever. Fix crawler access and off-site mentions first.

Do I need to rank on Google to be cited by ChatGPT?

No. In Ahrefs’ analysis of ChatGPT’s 1,000 most-cited pages, 28.3% had zero organic search visibility. Google rankings correlate with citations but are clearly not required. Domain authority, recency and being the kind of page a model wants to quote matter more than your position in the SERP.

Does schema markup improve SEO for ChatGPT?

It helps indirectly rather than as a direct ranking input. Schema makes your entities, prices and relationships unambiguous, which lowers the cost of resolving who you are. Nobody has demonstrated that adding FAQPage markup alone causes citations. Ship it because it is cheap and useful, not because it is a shortcut.

Why does ChatGPT cite Reddit so much?

Because Reddit threads contain first-person comparative opinion at enormous scale, which is exactly what conversational questions call for. As of July 2026 Reddit held about a 16.7% mention share of ChatGPT citations, nearly double Wikipedia’s. That share has proven volatile, so treat Reddit as one channel among several rather than the whole plan.

Can I pay to appear in ChatGPT answers?

Not in the organic citations. OpenAI stated at the launch of Instant Checkout that product results are organic and unsponsored, ranked purely on relevance. OpenAI does operate an ads surface with its own OAI-AdsBot crawler for validating submitted ad pages, but that is separate from the sources cited in a normal answer.

Does ChatGPT read JavaScript-rendered content?

No. OpenAI’s fetcher does not execute JavaScript, so anything injected client-side is invisible to it. Pages larger than 4 MB are also rejected outright with an HTTP 400 and nothing is read. Server-render your key answers and keep templates light, or the rest of your optimization work never gets evaluated.

How do I get my products to show in ChatGPT shopping?

Through a product feed, not through your website’s SEO. Merchants submit a regularly refreshed CSV or JSON feed with identifiers, pricing, inventory, media and fulfilment data, starting with a sample feed for validation and then daily snapshots. Access runs through chatgpt.com/merchants and is still partner-gated as of August 2026.

Is SEO for ChatGPT different from optimizing for Google AI Overviews?

The fundamentals overlap but the plumbing does not. AI Overviews draw on Google’s index and Googlebot, while ChatGPT draws mainly on OpenAI’s own index and OAI-SearchBot. Citation sources differ too. You need separate crawler permissions and separate tracking for each surface, even though the content work is largely shared.

Why did ChatGPT stop citing my site?

Three things break SEO for ChatGPT most often, so check them in order. First, whether a robots.txt, WAF or CDN rule started blocking OAI-SearchBot. Second, whether the page changed so the answer moved below the first 200 characters after the H1. Third, whether the citation share of a platform you depended on shifted, which happens without warning and without announcement.