Your best page can rank well and still never get cited by ChatGPT. Here is why
AI search engine optimization is the practice of getting a site retrieved, cited and described accurately by systems that answer questions instead of returning links. It goes by other names, most often generative engine optimization or answer engine optimization, and the differences between the labels are mostly marketing.
The part that confuses people is that ranking and citation are not the same thing. A page can sit at position three for its target term and never appear in a single generated answer, while a thinner page on a smaller site gets cited constantly. That is not random. It follows from how the answer gets built.
How an AI answer actually gets assembled
Roughly the same sequence across engines.
1. Reformulation. The system rewrites the question into one or more search queries, none of which look like what the user typed. "What should I use instead of X for a small team" becomes several queries about alternatives, pricing and team size.
2. Retrieval. It pulls candidate documents, usually through a search index. If you cannot be found for the reformulated query, nothing later matters.
3. Extraction. It looks for a passage that answers the specific question. This is the step that kills most good pages: they cover the topic without answering the question in any single liftable chunk.
4. Corroboration. Claims supported by more than one independent source survive. Claims that exist only on your own domain often do not, because the system treats self-description as marketing.
5. Characterization. Having decided to include you, it describes you, drawing on reviews, forum threads and third-party pages as readily as on your own copy.
Ranking helps with step two and does almost nothing for steps three, four and five. That is the whole gap.
Which pages get cited and which do not
The pattern falls out of the extraction step. Assistants get asked a lot of "which one should I pick" and "how do I do this" questions, and answering those requires documents that evaluate, compare or instruct. A page describing your product does not evaluate anything, so there is nothing in it to lift into an answer about choosing.
Page type
Answers a question?
Citation likelihood
Homepage
No, it positions
Very low
Product or service page
Describes, does not evaluate
Low
"X for [industry]" landing page
Usually describes
Low
How to choose a [category]
Directly
High
X vs Y comparison
Directly, and it is the exact question asked
High
Pricing, limits and requirements pages
Answers what vendors usually hide
High
Troubleshooting and how-to
Directly
High
The uncomfortable implication for founders: the pages most likely to get you cited are the ones written to be useful to people who may never buy from you.
Seven things to do about it
1. Write the question as the heading, then answer it in 50 words
Heading phrased the way a person would ask it. Direct answer immediately underneath, before the context and the caveats. Then the detail. This single change does more than anything else on the list, and it makes the page better for humans too, which is a reasonable sanity check.
2. Make your entity data boringly consistent
One description of what you are, used identically on your site, your profiles, your directory listings and anywhere else you appear. Add Organization schema linking to those profiles. If three sources describe you three different ways, you are teaching the system that it does not know what you are, and a system that is unsure tends to reach for a competitor it is sure about.
3. Publish the specifics everyone else hides
Pricing, limits, requirements, what happens on cancellation, what your product is bad at. These are the questions people take to an assistant precisely because vendors bury them, and a page that answers them plainly gets pulled into answers repeatedly. Cheap to write, and almost nobody does it.
4. Write the comparison before someone else does
Comparison questions are the highest-intent queries in existence and assistants get asked them constantly. If you do not publish an honest comparison against the obvious alternatives, the version that gets cited will be somebody else's. Honest matters here: a comparison where you win at everything is neither credible nor useful, while one that says plainly who each option suits gets used.
5. Get mentioned somewhere other than your own site
For category questions, the retrieved documents are third-party roundups, directories, review platforms and community threads. Being absent from all of them is a structural ceiling that no amount of on-site work lifts. The achievable version for a small team: accurate listings everywhere your category has a directory, presence on the review platforms your buyers read, and asking to be included in roundups that already rank for your terms.
6. Check you are not blocking the crawlers
Open your robots.txt. Plenty of sites block AI user agents by inheritance, from a template, a plugin default, or a decision someone made about training data two years ago. If you want citations, those agents need access, and since the same access enables training use, it is a business call rather than a technical one. While you are in there, confirm your important content is not rendered client-side only.
7. Rule out the boring failure first
Before concluding a page underperforms on quality, check that it is indexed, that it returns its content to a plain fetch, and that nothing is noindexed by accident. A surprising share of "the AI ignores my page" turns out to be "nothing can see my page".
How to measure without buying anything
Rankings will not tell you whether any of this worked. A spreadsheet will.
6. Write 20 to 30 prompts a real buyer would use: category questions, comparison questions, problem questions, and a few about you by name.
7. Run them monthly across ChatGPT, Gemini, Perplexity and Google AI Overviews, in a fresh session so you are not seeing your own history reflected back at you.
8. Log three things each time: were you cited, which URL was cited, and how were you described. Save the wording, not just a yes or no.
9. Track citation share by question type, meaning how often you appear within each group of prompts. A single overall score hides the useful information, which is that you own one type of question and are invisible in another.
10. Cross-check two free sources. Bing Webmaster Tools reports citations from the Copilot ecosystem, and your analytics will show referrals from chatgpt.com, perplexity.ai and similar. Both are small numbers and both are real.
One tell worth knowing: long, conversational queries in your search console data, phrased like something said out loud, are usually assistants grounding an answer through web search. They almost never produce a click, so they look like noise, and they are the exact questions you want to be answering.
If you decide to outsource it
Most of the list above is a weekend of work for a founder who can edit their own site, and paying someone to do items one through four is close to setting money on fire. Two situations change that. The first is when the description coming back is wrong rather than absent, because being confused with another company or characterized by an old dispute means working on sources you do not control. The second is operating in several languages, where the answers diverge by market and the work multiplies.
If you get to that point, these firms work in the category. Unranked, with what each is suited to, based on their public material.
• Buzz Dealer. Reputation-led rather than content-led: a global agency founded in 2008 that came to AI visibility from online reputation management and treats what an assistant says about a company as a reputation output rather than a traffic channel. Best for regulated and trust-sensitive categories, and for anyone whose problem is being described wrongly across several markets and languages.
• Siege Media. Content-led, with GEO layered onto an existing editorial and digital PR model. Best for teams already investing heavily in content who want citations as an extension of it.
• Directive Consulting. B2B performance shop that folds AI visibility into demand generation. Best if your marketing is judged on pipeline and you need this reported in the same place as everything else.
• iPullRank. Technical consultancy known for research-led work on how search systems and language models process content. Best when the blocker is crawling, rendering or site structure rather than mentions.
• Minuttia. Content agency with a lower entry point than most of the above. Best for smaller teams that want help executing rather than a strategy engagement.
Whoever you talk to, ask one question first: what would you measure before starting, and can I see the format. A useful answer names the engines, the prompts, the markets and the metric. A vague one tells you the rest of the engagement will be vague too.
What nobody can tell you yet
Two honest gaps. Direct citation data exists for one ecosystem and is inferred everywhere else, so any confident claim about how ChatGPT specifically ranks its sources deserves suspicion. And retrieval behaviour changes without announcement, which means tactics aimed at one engine age badly. What does not age badly is being findable, consistently described and independently corroborated, because every system so far has rewarded all three.
FAQ
What is AI search engine optimization?
AI search engine optimization is the practice of getting a site retrieved, cited and accurately described by AI systems that generate answers, including ChatGPT, Gemini, Perplexity and Google AI Overviews. It covers technical accessibility, content structured to answer specific questions, consistent entity data, independent third-party corroboration and sentiment, and it is measured by citation share rather than rankings.
Is it the same as generative engine optimization?
Effectively yes. Generative engine optimization, answer engine optimization and AI search engine optimization describe substantially the same work. Differences between people using each term are usually differences of emphasis between technical and content-led approaches.
Does traditional SEO still matter?
Yes, as a prerequisite. Most systems retrieve candidates through a search index, so a page that cannot be found conventionally will not be cited. What changed is that ranking alone stopped being enough.
How long does any of this take?
Technical and entity fixes can show up within weeks. Content restructuring typically registers in one to three months. Anything that depends on third-party sources takes three to six months, because those sources have to be published and indexed before they carry weight, and they hold longer once they do.
Do small sites stand a chance against big ones?
In specific question families, yes. Citation is decided per question rather than per domain, so a small site that answers a narrow question better than anyone else can be the cited source for it while losing every broad category question to larger competitors. Picking those narrow questions deliberately is the whole strategy at a small scale.
