Most advice on this subject fails at the same point. It offers a checklist without saying where the checklist came from, and it treats three different systems as though they behave identically. They do not. Google’s AI features draw on its own Search index. ChatGPT combines external search providers with its own crawler. Perplexity runs its own real time retrieval. What gets you into one does not automatically get you into another.
This guide is a working sequence rather than a checklist, ordered so that the things which block everything else come first. If you want the strategic framing behind it, start with our guide to generative engine optimization, and for the selection mechanics read how AI search engines choose which websites to cite.
One Caveat Worth Stating Plainly
There is no official ChatGPT SEO checklist. OpenAI publishes eligibility requirements, not ranking factors, and states that placement is not guaranteed. Google publishes best practices and explicitly says there are no additional requirements to appear in AI Overviews or AI Mode beyond sound SEO. Perplexity publishes almost nothing about selection.
Everything below is either a documented requirement, a general quality practice the platforms endorse, or a reasoned recommendation clearly labelled as such. Nothing here guarantees a citation, and anyone telling you otherwise is selling something.
Step One: Confirm You Are Actually Eligible
This is where a surprising number of businesses lose before they start, and it takes an afternoon to check.
To appear in Google’s generative AI features, a page must be indexed and eligible to be shown in Google Search with a snippet. That means checking three things: the page is indexed in Search Console, it carries no noindex directive, and it is not restricted by nosnippet or an unnecessarily tight max-snippet value. Marketing teams sometimes inherit restrictive snippet settings from an old publisher agreement and never revisit them.
ChatGPT
Open your robots.txt file and look for OAI-SearchBot. If it is disallowed, your content will not be included in ChatGPT search summaries and snippets. This crawler is separate from GPTBot, which relates to model training, and from ChatGPT-User, which handles some user initiated actions. Those decisions are independent, so you can permit search visibility while declining training use.
Two details worth knowing. Changes to robots.txt take roughly a day to affect search eligibility. And if OpenAI obtains a URL from a third party search provider while the page itself is disallowed, it may still surface the link and page title, which you would need a noindex tag to prevent. That tag can only be read if the crawler is allowed to fetch the page, which is a small logical trap.
Perplexity and Others
Check your robots.txt and any firewall or CDN rules for blanket AI crawler blocks. Many sites added these in 2023 and 2024 without distinguishing between training and search. Cloudflare and similar services also offer one click AI bot blocking that operates above robots.txt, so check there too.
Do this audit first. Refining your phrasing on pages that cannot be fetched is wasted effort.
Step Two: Answer the Question Before You Explain It
Conventional article structure warms the reader up. Generative retrieval works better when the answer is present, complete and near the top.
This is not about writing badly for machines. It is the same principle behind a good news lead. Give the answer, then earn the reader’s attention with the reasoning.
Before:
“Brand guidelines are an important part of any organisation’s identity toolkit. Many businesses wonder how comprehensive theirs should be, and the answer varies depending on a range of organisational factors that we will explore throughout this article.”
After:
“Most UK businesses need brand guidelines running to between twelve and forty pages. A single product company can usually document logo usage, colour, typography and tone of voice in twelve to twenty pages. Organisations with multiple sub brands, international markets or regulated communications typically need thirty to forty, with separate sections for each market. Anything beyond eighty pages tends to go unread by the teams meant to apply it.”
The second version answers the question, contains attributable figures, and names the conditions that change the answer. It also happens to be more useful to a human reader, which is the point.
Step Three: Make Claims Specific Enough to Attribute
Generative answers are assembled from claims. Vague statements offer nothing to lift.
Run a page through this test. Take any sentence and ask whether a system could quote it as evidence for something. “Response times vary” fails. “We respond to support tickets within four working hours during UK business hours, and within one hour for enterprise accounts” passes.
Three habits help:
- Replace ranges you have not verified with figures you can defend. If you do not know, say what you do know and why the rest varies.
- Name the conditions. “Usually four to six weeks, longer if legal review is required” is more citable than “four to six weeks”, because it survives contact with a different scenario.
- Put the fact in one sentence. Information spread across a paragraph is harder to extract than the same information stated once, cleanly, and then discussed.
Step Four: Cover the Surrounding Questions
Google documents a technique it calls query fan out, where the system issues several concurrent related searches rather than one. Its published example expands a query about a weed filled lawn into searches about herbicides, chemical free removal and prevention.
The implication is that depth across a topic supports visibility on any individual question within it. A single article about brand guidelines competes less effectively than a connected set covering guidelines, tone of voice, asset management and governance.
One important limit. Google warns that creating separate content for every possible query variation, particularly for fan out queries, risks breaching its scaled content abuse policy. Cover the questions that genuinely deserve their own treatment. Do not generate a page per permutation.
Step Five: Show Who Wrote It and Why They Know
Name the author. Link to a page explaining their background. State the organisation behind the content and what it actually does.
This is not a citation lever in itself. It contributes to the broader picture of whether a source is worth relying on, which is covered properly in our article on what makes a website trustworthy to AI search engines. It is included here because it is a fifteen minute fix that most sites have never done.
Step Six: Remove Ambiguity About Entities
Systems need to resolve which organisation, product or person you are referring to. Help them.
State the full name early rather than relying on a pronoun or an abbreviation established three paragraphs ago. Where a term is ambiguous, disambiguate once. If you are writing about a product called Atlas, and three other products share that name, say “Atlas, the browser from OpenAI” the first time rather than assuming context carries.
Keep your own business details consistent across your site, Google Business Profile, LinkedIn and any directory listings. Conflicting information weakens recognition.
Step Seven: Reference Your Sources and Link Out
Where you make a factual claim drawn from elsewhere, say where it came from and link to the primary source rather than a summary of it. Where you make a claim from your own work, say so and explain the basis.
Two reasons. First, it makes the claim verifiable, and corroborated claims are safer material for a system generating an answer. Second, it is simply better editorial practice, and readers can tell the difference between an article with sources and one without.
Step Eight: Use FAQs Where They Earn Their Place
A short FAQ addressing genuine follow up questions is useful. A twenty question FAQ stuffed with keyword variants is not, and creates exactly the kind of thin, repetitive content that quality systems are designed to discount.
The test is whether a real reader would ask the question after finishing the article. If not, cut it.
Note that Google states structured data is not required for its generative AI features and no special schema is needed. FAQ markup remains worth using for its own reasons in Search, but do not add it expecting an AI visibility effect.
Step Nine: Update Properly, Not Cosmetically
Where a topic changes, currency matters. Where it does not, republishing with a new date changes nothing except your credibility.
A meaningful update revises figures that have moved, removes guidance that is now wrong, replaces dead references, and reflects changes in the products or regulations discussed. Note what changed and when. If nothing has changed, leave it alone.
Step Ten: Sort the Technical Basics
Google’s guidance is unambiguous that its generative features depend on the same crawling and indexing infrastructure as Search, so the usual work applies: reasonable page speed, content that renders without requiring JavaScript execution to be readable, mobile usability, reduced duplicate content, and main content that is clearly distinguishable from navigation and adverts.
Semantic HTML is worth using because it helps screen readers and browser agents parse a page, though Google notes perfect markup is not required.
What to Avoid
Google’s own guidance names several popular tactics as unnecessary for its systems:
- llms.txt files and similar machine readable formats. Google states it does not use them, and that having one neither helps nor harms visibility in Google Search. Other services may use them, so maintaining one is a preference rather than a priority.
- Chunking content into small blocks. Explicitly described as unnecessary, with no ideal page length.
- Rewriting content specifically for AI systems. These systems understand synonyms and meaning, so you do not need to capture every phrasing variant.
- Pursuing inauthentic mentions. Named directly as a tactic to ignore.
- Overfocusing on structured data. Useful for rich results, not required for generative features.
Add one more from practice: do not restructure genuinely good long form content into stilted question and answer blocks because a tool told you to. The engagement cost is real and the benefit is unproven.
A Worked Example
Here is a paragraph of the kind that appears on thousands of agency websites, and a rewrite that gives a generative system something to work with.
Original:
“Our SEO services are designed to help your business grow. We take a bespoke approach tailored to your needs, working closely with you to deliver results that matter. Our experienced team uses the latest techniques to improve your visibility online.”
Rewritten:
“We run SEO for UK businesses turning over between two and twenty million pounds, typically in professional services, SaaS and specialist retail. Engagements start with a four week technical and content audit, followed by a rolling monthly programme covering technical fixes, content production and digital PR. Most clients see meaningful movement on non branded queries between four and seven months in, depending on domain age and competition. We do not take on clients who want guaranteed rankings, because nobody can offer them.”
The rewrite contains a defined audience, a described process, a stated timeframe with a condition attached, and an explicit exclusion. Each of those is an attributable claim. The original contains none.
A Sensible Order of Work
- Audit crawler access, indexing and snippet directives across the site.
- Identify your twenty most commercially important pages.
- Rewrite the opening two hundred words of each so the central question is answered directly.
- Replace vague claims with specific, defensible ones.
- Add author identification and organisational transparency.
- Fill the obvious gaps in topic coverage, one cluster at a time.
- Set up whatever measurement is available and review it monthly, as covered in our guide to measuring AI search visibility.
That sequence puts the blocking issues first and the slow compounding work last, which is the opposite of how most GEO projects are run.
Frequently Asked Questions
Is there an official checklist for ranking in ChatGPT?
No. OpenAI publishes eligibility requirements, principally that OAI-SearchBot must not be blocked, and states that results are ranked using multiple factors with placement not guaranteed. It does not publish those factors, and no third party has access to them.
Should I write in a question and answer format throughout?
Only where it suits the content. Answering the central question early is genuinely useful. Converting an entire article into a list of headings phrased as questions usually reads poorly and is not something the platforms ask for.
How long should content be for AI search?
Google states there is no ideal page length and that shorter or longer pages can both work depending on audience and subject. Length is not a lever. Specificity is.
Do I need different content for Google, ChatGPT and Perplexity?
No. The content work is largely shared. What differs is eligibility, which is technical, and off site presence, which matters more on platforms that lean heavily on third party sources.
Will these changes help my normal search rankings too?
Most of them should, because they amount to clearer, more specific, better evidenced content on a technically sound site. That overlap is precisely why Google frames this work as ordinary SEO rather than a separate discipline.
Published by BrandingX UK.
