How to Rank in ChatGPT

I started doing SEO in 2004. Florida, Panda, Penguin, Hummingbird, and a dozen core updates rearranged this industry, but none of them changed searcher behavior itself. ChatGPT did. OpenAI reported 900 million weekly active users in February 2026, more than double the year before, and a growing share of buying research now starts in a chat window instead of a results page.

Here is what most guides get wrong: getting cited is not a lottery. ChatGPT selects a small set of sources for every answer, and the selection follows patterns you can engineer for. This is the playbook I run on client sites.

How ChatGPT actually picks its sources

It decides across two layers, and the tactics for each are different. Layer one is the model itself: the training corpus determines whether ChatGPT knows your brand exists and what it believes about you. Layer two is live retrieval: when search activates, the system queries a search index (Bing is the dominant one), retrieves a candidate set, and a reranking step selects the passages worth citing. OpenAI's own crawler for this, OAI-SearchBot, fetches and reads the pages.

The practical consequence: if a page is not indexed and competitive in Bing, ChatGPT usually cannot cite it, no matter how good the content is. And visibility in AI answers is close to binary. You are in the cited set or you do not exist. There is no position 8 consolation prize.

01

Step 1: Open the door to the right crawlers

OpenAI runs three user agents, and confusing them is the most common self-inflicted wound I see in audits. GPTBot is the training crawler: blocking it only opts your content out of model training. OAI-SearchBot powers ChatGPT search: blocking it removes your pages from live citations. ChatGPT-User fetches pages a user asks ChatGPT to visit and is not used to determine search eligibility.

These are independent decisions. You can decline training and still allow citations, or the reverse. A Q1 2026 audit found that 41% of B2B sites still block at least one major AI crawler, usually a leftover rule from the 2023 blocking panic. Check your robots.txt before you touch anything else. It is the cheapest fix in this entire post, and OpenAI says changes take roughly 24 hours to register with its systems.

A sane baseline:

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: GPTBot
Allow: /

Swap the GPTBot line to Disallow if your policy is no training. Keep admin, checkout, and account paths blocked for every agent.

02

Step 2: Treat Bing like it matters, because now it does

Bing Webmaster Tools was the account everyone created once and never opened again. ChatGPT changed the math: its retrieval layer leans on Bing's index, which makes Bing indexation the entry ticket to citations. Verify your site, submit your sitemap, and fix Bing-specific crawl errors, which are not always the errors Google reports. Then benchmark your seed queries in Bing, not Google, because Google positions are necessary context but not sufficient. Enable IndexNow while you are in there so URL changes push to Bing instantly instead of waiting on a crawl.

03

Step 3: Lead with the answer

The reranking step selects passages, not pages. ChatGPT quotes sentences. Write your H2s as the questions buyers actually ask, then make the first 40-60 words under each heading answer that question outright, before any wind-up. Here is the test I use: lift any key sentence out of the page. If it still makes sense standing alone with your brand attached, it is citable. If it depends on three paragraphs of setup, the model will pass and quote someone who got to the point.

04

Step 4: Raise your fact density

Language models prefer specific claims over marketing language, because specific claims are quotable and defensible. Numbers, dates, named tools, thresholds, and original data all raise the odds a passage gets selected. The strongest version is publishing something only you can publish. First-party data, even from a small sample, makes you the primary source, and primary sources get cited by everyone downstream, including the models.

05

Step 5: Keep your dates honest

Cited URLs skew heavily toward pages with a recent, genuine dateModified. Freshness is a real selection signal in AI retrieval, more than it ever was in classic Google. The catch: bumping dates without changing anything earns you the trust profile of a spammer. My cadence is a true quarterly refresh on money pages. Update the stats, replace dead examples, tighten the answer blocks, and let the modified date reflect work that actually happened.

06

Step 6: Ship schema machines can parse

Schema markup does not rescue weak content, but it removes ambiguity about who wrote what and when, and cited pages disproportionately carry clean structured data. BlogPosting on posts, FAQPage where you genuinely answer questions, and Organization or Person markup with a complete sameAs graph so the model can connect your site to your profiles.

One thing I insist on: server-render every head tag and every line of schema. I learned this the expensive way. A client-side script once injected a noindex tag on my own site that only existed in the rendered DOM, and the site was de-indexed before I caught it. The full story is in my SEO audit checklist. AI crawlers are even less forgiving than Googlebot here, because most of them do not execute JavaScript at all.

07

Step 7: Make your entity impossible to confuse

Models hedge when facts conflict, and hedging models leave you out of the answer. Your name, description, location, and core claims should be identical across your site, LinkedIn, Google Business Profile, directories, and anywhere else you appear. Tie it all together with the sameAs graph in your schema. If your brand qualifies for a Wikidata entry, that is one of the highest-leverage hours you can spend, because structured entity databases are disproportionately represented in training data.

08

Step 8: Earn mentions where the model reads

Third-party mentions carry serious weight in both layers. Editorial coverage, industry roundups, comparison posts, and community threads teach the model that your brand belongs in the answer for your category. Backlinks still matter, but multiple 2026 analyses found they explain only a small share of which sources actually get cited. A mention without a link still counts here, which is a genuine break from classic SEO. Extractability and recognition do the heavy lifting now.

09

Step 9: Measure with a frozen prompt set

There are no rankings to check, so build the replacement. Write 30-50 prompts your real buyers would type, freeze the set, and run it on a fixed monthly cadence. Track two numbers: citation rate, meaning the percentage of prompts where you appear as a source, and share of voice against named competitors.

Then close the loop. For every prompt where a competitor is cited and you are not, open their cited page and ask one question: what makes this more extractable than mine? The answer is usually visible in thirty seconds.

What does not work

Keyword stuffing rephrased as prompt optimization. Buying AI SEO link packages. Treating llms.txt as a magic switch: it costs ten minutes to add and I do, but no major engine has confirmed using it for ranking, so it is hygiene, not strategy. And blocking every crawler while expecting to control your AI presence anyway: a March 2026 study of 4 million AI citations found that blocking crawlers does not reliably remove you from answers, because models keep learning about you from third-party sources. You do not get to opt out of the conversation. You only get to decide whether you are the primary source in it.

Frequently asked questions

How long does it take to get cited by ChatGPT?

Most sites that fix crawler access, Bing indexation, and answer-first structure see first citations within 60-90 days on lower-competition prompts. Competitive comparison prompts take longer, because third-party mentions have to accumulate before the model trusts you in that set.

Do I need to rank number one in Google to show up in ChatGPT?

No. ChatGPT's live retrieval leans on Bing, not Google. Strong Google rankings usually correlate, but I have seen pages with modest Google positions get cited because they were the most extractable source in Bing's candidate set.

Does blocking GPTBot hurt my ChatGPT visibility?

Blocking GPTBot only opts your content out of model training. Search citations run through OAI-SearchBot, a separate crawler with a separate robots.txt entry. Blocking both makes you invisible at the retrieval layer and leaves your story to third parties.

What should I track instead of rankings?

A frozen set of 30-50 buyer prompts, run on a fixed monthly cadence. Track citation rate, share of voice against named competitors, and AI-referred sessions in your analytics.