You ask ChatGPT for a service like yours. It names three businesses, links two of them, and none of them is you. Your Google rankings look fine. The answer box does not care.
That is the gap these guides cover. Classic search gave you ten blue links and a position to fight for. AI search gives the reader one written answer with a handful of sources inside it. Either your page is one of those sources, or the reader never hears of you.
The short answer
There is no secret file and no trick prompt. The engines that publish their rules say roughly the same thing: let their crawler in, be indexable, write pages that answer a question plainly and can be quoted, and keep them current. Google states there are no extra requirements for AI Overviews or AI Mode beyond being indexed and eligible for a snippet (Google Search Central). OpenAI says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers (OpenAI crawler docs). Perplexity asks you to allow PerplexityBot if you want to appear in its results (Perplexity docs).
Old flow → new flow:
keyword → ranking → click becomes question → answer → cited source → click (sometimes).
What AI search actually is, and how it differs from the ten blue links
Classic search handed you a list and let you choose. You typed a phrase, got ten links, and did the sorting yourself: you opened three, compared them, and picked one. The engine’s job ended at the list. Your job was to be high enough on it to get the click.
An AI answer does the sorting for the reader. The engine reads a question in plain language, pulls a small set of pages it already trusts on the topic, and writes one paragraph back. Sometimes it names a few of those pages as sources; sometimes it just states the fact and moves on. The reader often never leaves the chat. The list is gone, and so is the position you used to fight for.
Google is the useful halfway case, because its AI features sit inside the search engine you already know. AI Overviews and AI Mode use the same crawling and indexing as regular results, and Google says a page needs no extra treatment to be eligible beyond being indexed and snippet-eligible (Google Search Central). ChatGPT and Perplexity are standalone answer engines with their own crawlers, and they document far less about how they choose sources. The honest position across all three: a page they cannot read cannot be quoted, and a page nothing else on the web mentions is hard for any of them to trust.
Laid side by side, the shift is less mysterious than it sounds:
| Dimension | Classic SERP | AI answer |
|---|---|---|
| What the reader sees | A ranked list of links | One written answer, a few sources inside it |
| Who does the sorting | The reader, across ten tabs | The engine, before the reader sees anything |
| What you compete for | A position on the list | Being one of the handful of pages quoted |
| How many winners | Ten per page, plus more pages | Often two or three named sources |
| Where the click comes from | Your link in the list | A citation the reader chooses to follow |
| How you measure it | Rank and impressions | Whether you are named, and how often |
Notice the bottom row. In classic search you could watch a number move every day. In AI search the thing you are chasing is softer: did the answer mention you, and is it doing so more this month than last. That difference shapes everything below.
What “being cited by an AI answer” means
It helps to be precise, because people mean three different things by “showing up in ChatGPT” and only two of them are real.
The first is being named: the answer says your brand out loud, even without a link. That still sends a buyer looking for you by name, so it counts.
The second is being cited: the answer carries a clickable source that points to your page. That is the closest thing to the old click, and it is the one most people picture.
The third is the fantasy: being “ranked number one in ChatGPT.” There is no number one. An answer is not a list, and it changes between runs and between users. Chasing a rank that does not exist is how people end up buying files and prompts that do nothing.
So the goal is plain. You want the engine to reach for your page when it writes an answer in your field, and you want your name attached to the claim when it does. Everything an honest practitioner can do is aimed at removing the reasons an engine would skip you. None of it forces the engine’s hand.
The honest levers, and the ceiling on each
Here is what I actually have my agent work on, in order of how much it tends to matter, with the honest ceiling on each. None of these is a switch that turns on citations. Each one removes a reason to be passed over.
Entity clarity and sameAs
An engine answering “who does X” is not reading one page of yours. It is drawing on everything it has seen about your brand: your about page, your profiles, directory listings, mentions in other people’s articles. That bundle of facts is your entity, and the cleaner it is, the easier you are to name.
The work is slow and unglamorous: spell your name the same way everywhere, give your about page real facts instead of adjectives, and connect your official profiles to your site and back so the pieces line up. On the page, sameAs links in your Organization structured data are how you tell a machine “these profiles are also me.” It does not guarantee anything; it just removes the ambiguity that makes an engine hedge. The ceiling: a tidy entity makes you easy to describe, but you still have to be worth describing.
Citable pages that answer the question up top
An engine summarises. If a page cannot be reduced to one clear, specific claim with your name next to it, a page that can wins instead. So the pattern is boring and effective: put the direct answer in the first sentence or two under a clear heading, then the detail. One page, one job. Specific nouns over adjectives, because “restores the URL structure and 301-maps the rest” is quotable and “best-in-class service” is noise.
This is the lever with the highest ceiling, because it is also just good writing for humans. The guide on how to rank in ChatGPT breaks the citable-page pattern down line by line.
Structured data and llms.txt, the experimental signals
Structured data (an Organization, a Service, a FAQ) helps a machine read what a page is about. Use it because it describes the page accurately. Google is explicit that it is not a ticket into AI features, so treat it as plumbing, not magic.
llms.txt is the experimental one. It is a proposed convention, published at llmstxt.org, for a plain Markdown file that hands a model a short, curated map of your best content. It is cheap and harmless and worth an hour. But as of this writing none of Google, OpenAI or Perplexity documents reading a site’s llms.txt to decide what it cites. So the honest ceiling is low: it is a good habit, not a lever. If someone is selling it to you as the key to AI visibility, read the detail on what is llms.txt before you pay.
Freshness, so the answer does not quote a stale you
Answers about tools, prices and policies go stale fast, and an engine that notices a page has not moved in three years has a reason to prefer one updated last month. Keeping your key pages current is not a growth hack; it is maintenance that stops you decaying out of the answer. A real update changes the facts, not just the date. The deeper method lives in the content refresh guide.
How to measure whether any of it worked
You cannot improve what you do not check, and AI visibility is harder to check than rank, because answers vary run to run and user to user. One screenshot proves nothing. A routine proves a trend. Here is the one I run, monthly:
- Write down 10 to 20 real questions a buyer would ask, in their own words, not keyword phrases.
- Ask each one in ChatGPT, Perplexity and Google (watching for an AI Overview) on a fixed schedule.
- For each, record two things: were you named, and were you cited with a link. Note which pages were quoted instead of you.
- Divide by the number of questions and you have a rough mention rate and citation rate for the month. The absolute number matters less than the direction over several months.
- Read the pages that beat you. They are your brief for the next round.
For Google, there is one hard number you can trust: clicks from AI Overviews and AI Mode are counted inside the normal Search Console Performance report under the “Web” search type, not in a separate report (Google Search Central). For ChatGPT and Perplexity, watch referral traffic by source in your analytics, and treat it as a trend line rather than a precise count. This is deliberately a monthly check, not a dashboard that updates every hour, because the thing being measured simply does not move that fast.
The honest limits
I would rather lose the sale than oversell this, so here is the part most pages leave out.
Generative engine optimisation is experimental. The two standalone engines that matter most, ChatGPT and Perplexity, do not publish how they pick sources, so a lot of the work is informed by what they do document about crawling plus plain reasoning about what a summariser needs. No one outside those companies can promise a citation, and anyone who does is selling you a certainty they do not own.
Answers are also unstable. The same question can return different sources on Tuesday than it did on Monday, with nothing changed on your end. That is why the honest unit is a mention rate tracked over months, not a rank you refresh daily. It is why there is no “position one” to buy.
And the levers are real but bounded. A clean entity, quotable pages, structured data, an llms.txt and fresh content together remove the reasons an engine would skip you. They do not reach into the model and force a name. The honest promise is the work and the log of what changed, never the outcome. If an external fact here ever drifts from what a vendor currently documents, treat the vendor’s own page as the source of truth, not this one.
The guides, and who each one is for
| Guide | Read it if | What you leave with |
|---|---|---|
| How to rank in ChatGPT and AI answers | You want your brand named in AI answers | The crawler checklist, citable-page pattern, and how to track mentions |
| What is llms.txt | Someone told you to add an llms.txt file | What the proposal says, who documents using it, and whether it is worth an hour |
| Content refresh | Old posts are slowly losing clicks | A way to find decaying pages and update them for real, not just the date |
| Is AI content bad for SEO? | You publish with AI help and worry about Google | What Google actually says, and where the scaled content abuse line sits |
Start with the guide on how to rank in ChatGPT if you only read one; it ties the rest together. If a consultant has already sold you on a magic file, read our answer to “What is llms.txt?” before you pay for it. If your traffic chart looks like a slow leak rather than a cliff, the fix is usually a content refresh, not more new posts. And if you are asking “Is AI content bad for SEO?” the honest answer is about quality and intent, not about who held the pen.
How these guides were put together
Every claim about how an engine behaves is checked against that engine’s own documentation and linked: Google Search Central, OpenAI’s crawler page, Perplexity’s bot page, and the llms.txt proposal itself. Where a vendor does not document something, the guide says so rather than guessing. No ranking tests or visibility numbers are claimed, because none have been run for publication yet. The method is written up in full on the methodology page.
One disclosure, since it is the topic: this site was built end-to-end by the same AI agent pipeline that does the work we sell, and a human reviews every page before it goes live.
If you would rather hand it over
Two separate services cover this ground. Our GEO services work on getting your brand named in ChatGPT, Perplexity and Google AI answers: entity signals, citable pages, llms.txt plus structured data, and a monthly mention check. Autopilot is done-for-you autoblogging: new posts on keywords people search with intent to visit, published on a schedule, checked before they go live, with a Search Console report. They can be bundled; see pricing. Neither comes with a ranking promise. Nobody honest can promise what a language model will say tomorrow.
FAQ
Is AI search replacing Google?
Not replacing; it is now part of Google itself. AI Overviews and AI Mode sit inside Google Search, and Google reports their traffic inside the normal Search Console Performance report under the “Web” search type (source).
Do I need special schema or files for AI answers?
Google says no: no new machine-readable files, AI text files or special schema.org markup are needed for its AI features. Structured data still helps machines read your page; it is just not a ticket in.
Should I block AI crawlers?
Only the ones you mean to. OpenAI separates its search crawler (OAI-SearchBot) from its training crawler (GPTBot), and each robots.txt rule is independent. Blocking training is a choice; blocking search means you are not in the answer.
Does llms.txt actually work?
It works as a tidy map that an agent already pointed at your site can fetch. It does not work as a ranking signal: as of this writing none of Google, OpenAI or Perplexity documents reading a site’s llms.txt to choose what it cites (llmstxt.org is the proposal itself, and the AI labs publish one for their own docs, which is not the same as reading yours). Make one if it takes an hour; do not pay for it as a standalone line item.
Can you guarantee ChatGPT will cite my site?
No, and anyone who says yes is guessing or selling. The engines that decide do not publish how they pick sources, and the same question can return different sources from one day to the next. What the work can do is remove every documented reason to skip you: let the crawlers in, be indexable, write pages that can be quoted, and keep your brand facts consistent. The outcome belongs to the engine. The promise is the work and the log of what changed.
Is AI-written content bad for SEO?
Short version: no, not by itself. Google judges the page by quality and intent, not by who or what held the pen, and the line it enforces is scaled content abuse, not “a machine helped.” The full detail, with Google’s own wording, is in the guide linked above under the table. The same rule applies to an llms.txt or an about page drafted with AI: check every line, because a fact a stranger can repeat is what you are after.
The answer box is the new front page. You still have to be worth quoting.