Getting cited when the answer replaces the click.
A growing share of buying research now ends inside an AI answer rather than on your website. This is the work of being the source that answer is built from, across AI Overviews, ChatGPT, Perplexity and Gemini.
Free AI visibility report
We run a set of real buying questions from your category across six AI surfaces and send you what they say, who they cite, and whether you appear at all.
Three acronyms, and the industry has not settled on them.
You will also see AI SEO and LLMO used for the same work. What matters is not the label but which of these three problems you actually have, because the answer changes what we do.
SEO
Search Engine Optimization
Ranking in the list of blue links. Still the foundation, and still where most qualified traffic comes from for the majority of businesses.
AEO
Answer Engine Optimization
Being the source a direct answer is pulled from. Featured snippets, knowledge panels and the summary at the top of an AI Overview.
GEO
Generative Engine Optimization
Being cited and recommended when a model synthesises an answer from many sources. Broader than any one page, involving what the model knows about you as an entity.
The click is no longer guaranteed to follow the research.
The answer is the destination
Someone asks a question, gets a synthesised answer naming three vendors, and contacts one. If you were not named, you were never in the running and you see nothing in analytics.
Your crawlers may be blocked
Many sites block AI crawlers without knowing it, sometimes through a CDN default rather than a deliberate choice. If the bots cannot read you, nothing else matters.
Most citations are not from your site
A large share of what AI systems reference comes from third party sources: directories, reviews, forums and press. Optimising only your own pages leaves most of the surface untouched.
Nobody is measuring it
None of this appears in Google Analytics. Without deliberate tracking you have no idea whether you are recommended, ignored, or described inaccurately.
Most GEO advice has never been tested. Some of it has.
This field generates a lot of confident guidance from people selling tools. Three published studies are worth knowing about, and they shape what we will and will not spend your budget on.
Evidence markers work
The foundational GEO study ran 10k queries testing content changes against whether an answer cited the page. Adding machine extractable provenance produced the biggest gains: direct quotations, statistics and citations, each worth roughly 25 to 40% more visibility.
Most tactics do not help
The first systematic benchmark of conversational SEO techniques reached a blunt conclusion: the majority of popular tactics produce no benefit, several actively hurt, and plain source relevance keeps working.
It is a pipeline, not a score
A critical survey reviewing 45 studies found that generative visibility is not one ranking factor but a chain of stages, and that winning one stage does not guarantee the next.
Seven stages between your page and a citation.
Visibility in an AI answer is not a position you hold. It is a sequence you have to survive. Each stage has different failure modes, and the diagnosis determines the work.
Engine decides to search
Some answers come from the model alone and never touch the live web at all
Crawl
Your robots file, CDN rules and rendering decide whether bots can read you
Index
The page has to be stored somewhere the retrieval step can reach it
Retrieve
Queries get broken into sub questions, and you compete on each one separately
Context allocation
Retrieved does not mean used. Space in the context window is limited
Citation
Whether the answer credits you with a link, a mention, or nothing at all
Absorption
Whether the substance of your view survives into the answer, accurately
Supported by evidence or by Google directly
- Making sure AI crawlers can actually reach and render your pages
- Adding statistics, quotations and cited sources to key pages
- Structured data using supported vocabulary that matches visible content
- Entity clarity, so the model knows what you are and what you are not
- Third party presence: reviews, directories, press, comparison sites
- Ordinary strong SEO, which still feeds the generative surfaces
Untested, disproven, or explicitly ignored
- llms.txt files sold as a ranking factor, covered in detail below
- Hidden text or instructions aimed at manipulating a model
- Prompt engineering tricks embedded in page copy
- Bulk publishing AI content to increase surface area
- Any 'special markup' claimed to guarantee AI Overview inclusion
- Guaranteed citation rates, which nobody can honestly offer
Find out what the models say about you right now.
We run a set of real buying questions from your category across six AI surfaces and send you the transcripts: whether you are named, who is named instead, whether the description is accurate, and which sources each answer was built from.
Twelve workstreams, grouped by which stage they fix.
The audit tells us which stage you are failing at. There is no point writing evidence rich content if a CDN rule is blocking the crawler that would read it.
Access, stages one to three
AI crawler access audit
Checking whether GPTBot, ClaudeBot, PerplexityBot and the rest can actually reach your pages. A surprising number of sites block them by accident.
Rendering and delivery
AI crawlers largely read the HTML your server returns. Content that only appears after JavaScript executes may as well not exist for them.
Structured data
Supported schema vocabulary that matches what is visibly on the page, used to clarify what your entities are rather than to decorate the source code.
Content and evidence, stages four to six
Evidence enrichment
The single best supported tactic in the literature. Adding verifiable statistics, direct quotations and cited sources to the pages you most want cited.
Query fan-out mapping
AI systems break a question into sub questions and search each separately. We map the sub questions and make sure something of yours answers each one.
Extractable formatting
Clear definitions, direct answers near the top, self contained sections and unambiguous headings. Content that can be lifted cleanly without the surrounding page.
Question coverage
The literal questions buyers ask, answered directly on your site rather than implied across three paragraphs of positioning language.
Comparison and alternatives
Buying prompts are overwhelmingly comparative. If no honest comparison of you against your competitors exists, the model will build one from whatever does exist.
Narrative accuracy
Models sometimes describe businesses incorrectly: wrong market, discontinued product, outdated pricing. We find those errors and correct the sources feeding them.
Entity and authority, the part that is not on your website
Third party presence
A large majority of AI references come from sources other than the brand's own site. Directories, review platforms, industry lists and comparison sites do more work than your homepage.
Digital PR and citations
Press coverage, expert commentary and data led stories. Being quoted by sources the models already trust is more effective than asserting expertise on your own domain.
Entity consistency
Same name, same description, same category everywhere. Inconsistency across profiles gives a model conflicting information and it resolves that conflict unpredictably.
About llms.txt, which you have probably been pitched.
It is the most sold item in this category and the least supported. Many agencies sell the creation of an llms.txt file as an "optimization" that guarantees AI models will read your site properly. Currently, almost no major consumer LLM uses these files for crawling or indexing decisions. Focus on standard crawler access, rendering, and structured data instead.