How to Rank in ChatGPT Search: 7 Steps That Work in 2026
You cannot rank in ChatGPT search the way you rank on Google. There is no page of ten results and no position number. ChatGPT searches the web, reads a handful of pages and writes an answer from them. To appear, your page has to be reachable by OpenAI's search crawler, clear enough to be picked from the pool and specific enough to be the source the answer is built on.
The good news: almost none of that is new. It is the same work as good search optimization, scored differently. Here is the version I would hand to a team starting this week.
TL;DR
- ChatGPT does not rank pages. It retrieves a pool of candidates, quotes some of them and links a few.
- Inclusion starts with crawler access, not content. If
OAI-SearchBotis blocked, nothing else matters. - Pages that sit at the top of the retrieval pool are cited far more often, so Google visibility still helps.
- The pages that get cited answer one question precisely in the first two sentences under a heading.
- Being mentioned on sites you do not own is often worth more than another post on your own.
- Measure with the same set of prompts over time, not with a screenshot of one lucky answer.
What "ranking" means in ChatGPT search
When someone asks ChatGPT a question, the system does not consult one ordered list. It breaks the question into smaller searches, pulls candidate pages, reads their titles and snippets, opens some of them and then writes a response using a few sources.
That creates four separate outcomes, and they are not the same thing:
| Outcome | What happened | Why it matters |
|---|---|---|
| Retrieved | Your page made the candidate pool | Without this, nothing else can follow |
| Mentioned | Your brand name is in the answer text | Useful even when no link is shown |
| Cited | Your URL is listed as a source | This sends the traffic you can measure |
| Clicked | A person follows the link | Rarer than the first three |
Most advice collapses these into "visibility". A brand can be named in an answer without being cited, and a page can be cited without sending a single visit. The practical target is the third line: be one of the sources the answer is built from.
The rest of this guide is the work that gets you there, in the order I would do it.
Step 1: Make sure ChatGPT is allowed to crawl you
This is the least glamorous step and it fixes more sites than any content tactic.
OpenAI runs separate crawlers for separate jobs. OAI-SearchBot is the one that surfaces websites in ChatGPT's search features. GPTBot collects content for model training. They are independent settings. You can allow search and refuse training, or the reverse.
OpenAI's documentation says sites that block OAI-SearchBot will not appear in ChatGPT search answers (OpenAI crawler docs). If a developer added Disallow: / for every AI bot in 2023 and nobody revisited it, that site is not losing a ranking contest. It was never in one.
Three checks worth running this week:
robots.txt. Confirm OAI-SearchBot is allowed and not caught by a wildcard block meant for other bots.- The layer above
robots.txt. A firewall, CDN or bot-protection rule can return 403 to a crawler even when your file says yes. OpenAI publishes its crawler IP ranges so you can allow them explicitly. - Timing. OpenAI says changes take about 24 hours to apply. That is slow enough to feel broken and fast enough to test.
While you are there, submit your sitemap in Bing Webmaster Tools as well as Google Search Console. OpenAI has not published its full retrieval recipe, so being easy to crawl and index everywhere beats being clever about one pipeline.
For the longer explanation of how ChatGPT turns a question into a set of sources, see our guide to ChatGPT as a search engine.
Step 2: Stay in the pool that Google and ChatGPT share
Once access is fixed, the contest becomes retrieval. And retrieval still rewards ordinary search visibility.
AirOps analysed 354,000 pages across 16,851 queries and found that a page in the first position of ChatGPT's retrieval results was cited 58% of the time. By position ten, that fell to 14%. Pages that were never cited had a median retrieval rank of 13; pages that were cited consistently had a median rank of 2.5 (AirOps, The Fan-Out Effect).
Ranking gets you into the room. It does not guarantee you get quoted.
There is a useful nuance. A study of 19,556 queries found that Google's top three URLs predicted ChatGPT citations poorly at the page level, but domain-level alignment was much stronger (Zenodo, 2026). AI platforms tend to trust the same domains as Google, then pick different pages from them.
What that means in practice: keep the pages that already rank well, strengthen the domain around them, and stop treating AI search as a reason to abandon Google. Google itself says its AI features have no additional requirements beyond pages that are indexed and eligible for a snippet (Google Search Central).
If you need a refresh on the basics, our guide to getting seen on Google covers the crawl, index and content side in more detail.
Step 3: Write the answer before the context
Once you are in the pool, the contest changes from authority to fit.
ChatGPT breaks a question into sub-questions. If someone asks "how do I get my brand into ChatGPT answers", the smaller searches might be about crawler access, citation behavior, tools and measurement. A page that answers one of those precisely beats a broad page that gestures at all of them.
The data backs the intuition. In Ahrefs' analysis of 1.4 million ChatGPT prompts, cited pages had a title similarity score of 0.602 to the prompt, against 0.484 for pages that were retrieved but not cited (Ahrefs, April 2026). Headings that match the question beat headings that require interpretation.
What that looks like on the page:
- Answer in the first 50 to 150 words under a question-shaped heading. Then expand. If a reader or a model stops after two paragraphs, they should still have the answer.
- Use the words people actually type. Put the phrase in the title, one H2 and the opening paragraph, then write like a person.
- Go deep on one question, not wide on ten. AirOps found pages covering 26% to 50% of related subtopics outperformed pages that covered everything. The best-performing length in that dataset was roughly 500 to 2,000 words.
- Give the answer a shape. A numbered list, a short table or a definition is easier to lift than four paragraphs of context.
One warning: do not turn this into a template factory. Ten near-identical pages targeting ten phrasings of one question is the 2026 version of keyword stuffing, and it reads as thin to a model as it does to a person.
If you want help choosing which questions deserve a page, start with keyword research for AI search.
Step 4: Make your claims quotable
A model cites passages it can use without rewriting them. Vague claims are hard to use, so they get skipped.
The original generative engine optimization research, presented at KDD 2024, tested ways of making content more citable. Adding citations, quotations and statistics produced 30% to 40% relative gains in visibility across AI engines (Aggarwal et al., 2024).
That finding is easy to apply:
- Put one named source behind each major claim. A dated report, a study, a vendor document, a government page.
- Use numbers with dates. "58% of the time, per a 2026 study of 354,000 pages" is quotable. "Significantly better" is not.
- Define your terms once, plainly. Models reuse clean definitions.
- Keep paragraphs short enough to lift. One idea per paragraph, two to four sentences.
This is also where original data pays off. A benchmark, a survey or a small experiment of your own travels further than an opinion, because nobody else can quote it from somewhere else.
Step 5: Get mentioned where you do not control the page
Here is the part that makes on-page specialists uncomfortable. Much of what ChatGPT says about a category is assembled from pages you do not own.
Semrush analysed 126 million US AI search prompts from January to April 2026 and found that ChatGPT cites roughly 15 sources per response, leaning heavily on community and reference platforms such as Reddit and Wikipedia. Gemini cites about 3 (Semrush, 2026 AI Visibility Index).
Fifteen slots sounds generous until you notice how many go to the same reference sites and the same comparison pages. If your brand only exists on your own domain, you are competing for the leftovers.
The work here is old-fashioned and slow:
- Get into the roundups and comparison pages already cited in your category. Not by pitching a link, but by being worth including.
- Answer questions in the communities where your buyers actually ask them. A useful reply from a named person still beats a press release.
- Give journalists and bloggers something to quote. Data, a clear correction, a named expert.
- Make your own site easy to describe. A one-line description, an About page and a consistent product name.
For the deeper version of this argument, see our GEO guide.
Step 6: Keep your brand facts identical everywhere
Models resolve a brand by matching what several sources say about it. Contradictions get hedged, attributed to someone else, or dropped.
Pick one version of each fact and repeat it: the product name, the one-line description, the founding year, the audience, the pricing model. Use the same wording on your site, your LinkedIn page, your directory listings and any profiles you control. Add real author bios with names and roles. An "About" page that says plainly what the company does is worth more than a page of brand adjectives.
This is unglamorous entity work, and it is one of the few things you fully control.
Step 7: Measure with prompts, not screenshots
You cannot open a report that lists what people typed into ChatGPT. The substitute is a fixed set of prompts run on a schedule.
- Pick 20 to 40 prompts written the way a buyer would ask, including "X vs Y" and "is X worth it".
- Run them monthly in ChatGPT, and alongside Google AI Overviews, Perplexity and Gemini.
- Record four things per prompt: whether your brand is mentioned, whether your URL is cited, which competitors appear and whether the description is accurate.
From that record you get three numbers worth reporting: mention rate, citation rate and share of voice against competitors. Rankings do not exist on this surface, so do not pretend they do.
Two measurement notes that save people from wasted months:
- Engine results do not transfer. ChatGPT citing 15 sources per answer and Gemini citing 3 are different games. A brand can be strong in one and absent in the other.
- Bing's AI Performance report is not a ChatGPT report. Microsoft's dashboard, in public preview since February 2026, covers Copilot and Bing's AI surfaces, and Microsoft says the data does not indicate ranking or authority (Bing Webmaster Blog). It is a useful signal, not the whole picture.
For traffic, at least, attribution is clean. Links ChatGPT sends carry utm_source=chatgpt.com, so your analytics can separate AI referrals from organic search. It will be a small number for a while. Watch the trend, not the daily total.
If you want a starting baseline without building a spreadsheet, our free AI visibility checker tests whether AI crawlers can reach your site at all.
The 30-day version
| Week | What to do | Output |
|---|---|---|
| Days 1 to 3 | Audit robots.txt, firewall rules and sitemaps. Confirm OAI-SearchBot can reach you. |
A yes or no answer on access |
| Days 3 to 7 | Run 20 prompts across ChatGPT, Perplexity and Google AI Mode. Record mentions, citations and competitors. | A baseline you can compare against |
| Week 2 | Take the three prompts where competitors are cited and you are absent. Check whether you have a page that answers each one. | A list of missing pages |
| Week 3 | Write or fix those pages: answer first, evidence behind each claim, one question per page. | Two or three strong pages |
| Week 4 | List the five third-party pages cited most in your category. Find a legitimate reason to be useful to them. | One or two real relationships |
| Month 2 | Repeat the same prompts. | A trend, not an anecdote |
Five mistakes worth skipping
Blocking the wrong bot. Blocking GPTBot to opt out of training is a legitimate choice. Blocking OAI-SearchBot removes you from ChatGPT search. They are different decisions.
Assuming schema markup is the unlock. Google says its AI features need no special schema, and OpenAI's eligibility rule is crawler access. Structured data helps search understand a page, but it is not a citation switch.
Writing for the phrase instead of the question. A page that repeats "how to rank in ChatGPT search" thirty times answers nothing. One clear sentence that resolves the reader's question does more.
Chasing a visibility score. No AI engine publishes an official ranking. Every "score out of 100" is a vendor's estimate, and two tools can disagree on the same brand on the same day.
Measuring once. A screenshot of ChatGPT mentioning your brand is an anecdote. The same thirty prompts, run monthly, is a metric.
Frequently asked questions
Can you rank number one in ChatGPT search?
No. ChatGPT does not show a ranked list, so there is no number one to hold. It retrieves candidate pages and cites a few sources in a written answer. The closest equivalent is being cited consistently for the prompts that matter to your business.
Does ChatGPT use Google to search?
OpenAI has not published its full retrieval recipe, and the details have changed over time. What is public and stable is the part you can act on: OpenAI runs its own crawler, OAI-SearchBot, and access to that crawler decides whether your pages can appear. Ahrefs found that about 88% of the URLs ChatGPT cited came from its general web search index, so ordinary search visibility still feeds the pool.
How long does it take to appear in ChatGPT answers?
Crawler changes take about 24 hours to apply. A citation is not on a timer: it depends on what the model retrieves for a given prompt. Expect weeks to months of consistent work before a citation rate moves, and expect answers to vary between runs.
Do I need a special file for ChatGPT, like llms.txt?
No. Google's guidance says Search ignores llms.txt, and OpenAI's eligibility rule is crawler access. The file can be useful for developer documentation read by coding agents. It is not a visibility lever for ChatGPT search.
Does ChatGPT traffic show up in analytics?
Yes, if you look for it. Links from ChatGPT carry utm_source=chatgpt.com, so you can build a segment for AI referrals and compare it with organic search over time.
Key takeaways
- ChatGPT has no ranking positions. It retrieves pages and cites sources.
- Crawler access comes first. Without OAI-SearchBot, content work cannot pay off.
- Retrieval still follows search visibility: a strong pool position makes a citation much more likely.
- Answer one question precisely, early, with evidence a model can quote.
- Mentions on sites you do not own carry a large share of the sources in AI answers.
- Measure mention rate and citation rate on a fixed prompt set, monthly.
Sources
- OpenAI, crawler documentation
- Google Search Central, AI features and your website
- Ahrefs, Why ChatGPT Cites Pages
- AirOps, The Fan-Out Effect
- Zenodo, Google rank and AI citation
- Semrush, 2026 AI Visibility Index
- Aggarwal et al., Generative Engine Optimization, KDD 2024
- Bing Webmaster Blog, AI Performance public preview
HeyEmmett helps brands get found in ChatGPT, Claude and Google's AI answers. If you would rather see where you stand first, run the free AI visibility check and start with the pages that should already be winning those answers.