You type “best [your service] for small businesses” into ChatGPT. It writes a tidy paragraph, names three competitors, and cites a listicle you have never heard of. Your site, the one that ranks on page one of Google for the same phrase, is not in it.
So the question is not “how to rank in ChatGPT” in the old sense. There is no position seven in a chat answer. The real question is: how does a page become one of the few sources an AI engine reads, trusts and names when it writes its answer?
People call the work different things: generative engine optimization (GEO), answer engine optimization (AEO), AI search optimization, AI visibility. Same job, different badge. This guide covers what the engines themselves document, what you can do this week, and how to track whether any of it worked.
The shift: from ranking to being cited
Classic search and AI search reward overlapping things, but the unit of success changed.
Classic: keyword → ten links → you are link three → click.
AI answer: question → engine fetches a handful of pages → writes one answer → names a few sources → some readers click.
Think of it like the difference between a shop window on a busy street and a recommendation from a friend. The window gets seen by everyone walking past; the friend only names two or three shops, but the person listening already trusts the pick. You want to be the shop the friend remembers, and the friend only remembers shops that are easy to describe in one sentence.
That last part matters more than any tactic. An AI engine summarises. If your page cannot be summarised into a clear, specific claim with your name attached, it gets skipped for a page that can.
Step 1: let the right crawlers in
This is the boring step that quietly breaks more sites than anything else. Each engine uses its own crawler, and a blanket “block AI bots” rule copied from a forum can remove you from AI search entirely.
| Engine | Crawler for search answers | What the vendor says |
|---|---|---|
| ChatGPT search | OAI-SearchBot | Sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. GPTBot (training) is controlled separately. |
| Perplexity | PerplexityBot | Used to surface and link websites in Perplexity results; Perplexity recommends allowing it. Not used to train foundation models. |
| Google AI Overviews and AI Mode | Googlebot (normal Search crawling) | No extra requirements: the page must be indexed and eligible to show a snippet. |
Sources: OpenAI’s crawler documentation, Perplexity’s bot documentation, Google’s AI features documentation. OpenAI and Perplexity both note that robots.txt changes can take about a day to take effect.
Two details worth knowing. First, OpenAI’s rules are independent per bot, so you can block training (GPTBot) and still allow search (OAI-SearchBot). Second, user-triggered fetchers such as ChatGPT-User and Perplexity-User may not follow robots.txt, because a person asked for that page; both vendors say this in their docs. That is not a loophole for you, just a reason not to panic when you see them in your logs.
Check your robots.txt, your CDN or firewall bot rules, and any security plugin that blocks “AI scrapers” by default. A bot blocked at the firewall never even reads robots.txt.
Step 2: be indexable in normal search first
Google is explicit that AI Overviews and AI Mode use the same eligibility as regular results: indexed, crawlable, snippet-eligible, with the important content in text. If a page carries noindex or nosnippet, it is out of the AI features too. Google also says there is no special schema.org markup and no “AI text file” you need to add to appear.
For ChatGPT and Perplexity, the vendors do not publish a full list of how they pick sources. The guidance they do publish is about letting their search crawlers in: OAI-SearchBot for ChatGPT (OpenAI) and PerplexityBot for Perplexity (Perplexity). What is safe to say: a page they cannot crawl cannot be cited, and a page nobody links to or mentions is harder for any system to trust.
Step 3: write pages that can be quoted
This is where most of the actual work lives. A citable page has a few habits:
- The answer comes first. Put a one or two sentence direct answer under the question heading, then the detail. Engines lift clean statements; they skip throat-clearing.
- One page, one job. A page that answers “what does a site restore cost” should not also try to be your about page and your blog roundup.
- Specific nouns over adjectives. “Restores the URL structure and 301-maps the rest” is quotable. “Best-in-class restoration” is not.
- Your name next to the claim. If the brand only appears in the logo, a summary can lift the fact and drop the name.
- Sources you can point to. Link the primary document for any factual claim. It helps the reader, and it is the same habit that makes you trustworthy to a summariser.
- Kept current. Answers about tools and prices go stale fast. A page last touched years ago competes badly with one updated last month. More on that in our guide to content decay.
Structured data helps machines read what a page is about (an Organization, a Service, a FAQ). Use it because it describes the page accurately, not because it is a ticket into an answer; Google says it is not.
Step 4: build the entity, not just the page
An AI engine answering “who does X” is not only reading your page. It is drawing on everything it has seen about your brand: your about page, directory listings, profiles, reviews on third-party sites, mentions in articles. That body of facts is your entity.
The work here is plain and slow:
- Make your name, description and contact details consistent everywhere they appear.
- Give your about page real facts: what you do, for whom, how. No invented history.
- Link your official profiles from your site and back, so the pieces connect.
- Earn mentions on pages the engines already read for your topic: industry roundups, community answers, partner pages. Earn, not buy; a paid list that nobody reads helps nobody.
Old SEO: page → links → rank. AI visibility: entity → mentions → cited in the answer. The page still matters. It is just no longer the whole story.
Step 5: track AI visibility
You cannot improve what you do not check. AI visibility tracking is less precise than rank tracking, because answers change between runs and between users, but a simple routine works:
- Write down 10 to 20 real questions a buyer would ask, in their words.
- Ask each one in ChatGPT, Perplexity and Google (watching for an AI Overview) on a fixed schedule, monthly is enough.
- Record whether your brand is named, whether you are linked, and which pages are cited instead of you.
- Read the pages that beat you. They are your brief.
For Google, AI Overviews and AI Mode clicks are counted inside the normal Search Console Performance report under the “Web” search type, not in a separate report (Google). For ChatGPT, OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs, so those visits can be tracked in your analytics (OpenAI). For Perplexity, look at referral traffic in your analytics by source.
A worked example (hypothetical)
Say you run a small bookkeeping firm and want to be named when someone asks “bookkeeper for Etsy sellers.”
- Crawlers: your security plugin blocks “AI bots”. You allow OAI-SearchBot and PerplexityBot, keep GPTBot blocked because you do not want training use. Googlebot was never blocked.
- Page: your services page lists eight services in one paragraph. You split out a page titled for Etsy sellers, open it with a two-sentence answer of what you do for them and what it includes, and add a short FAQ with the questions you actually get asked.
- Entity: your business name is spelled two ways across profiles. You fix it, and update the about page with the facts.
- Tracking: you ask the same 12 questions monthly and log the results.
Whether that gets you named is up to the engines. What it does is remove every reason for them to skip you.
Common mistakes
- Blocking every AI bot by default. It protects you from training use and also removes you from AI search. Block deliberately.
- Buying a magic file. Some services sell llms.txt as the key to AI rankings. Google says it needs no AI text files for its AI features, and OpenAI and Perplexity do not document reading a site’s llms.txt to choose sources. Read what is llms.txt before you pay for one.
- Writing for the bot. Stuffing pages with “according to AI” phrasing or Q&A blocks nobody asked for. Engines summarise for people; write for people.
- Mass-producing thin pages. Google’s spam policies treat many low-value pages made to manipulate rankings as scaled content abuse, however they were produced.
- Measuring once. AI answers vary run to run. One screenshot proves nothing; a monthly log shows a trend.
When to hand it over
All of the above is doable by one patient person with a spreadsheet. If you would rather not, our AI search visibility service does the entity signals, the citable pages, llms.txt and structured data, and the monthly mention check. No promise that ChatGPT will name you, because no one controls that; just the work and the log of what changed.
The friend only recommends the shop she can describe in one sentence. Make sure yours has one.