What GEO means
Generative Engine Optimization (GEO) is the practice of making your business likely to be cited in answers produced by AI systems — ChatGPT, Perplexity, Claude, Google AI Overviews, Copilot — rather than merely ranked in a list of blue links.
The distinction matters because the interaction has changed shape. Traditional search returns ten results and the user chooses. A generative engine returns one synthesised answer, citing a handful of sources. If you are not among those sources, you are not on the page at all — there is no page two to be on.
Why this is worth your attention now
Two things are happening at once. Google now shows AI Overviews above the organic results for a large and growing share of queries, which pushes traditional results further down the page. Separately, a meaningful number of people have simply stopped starting at a search box and now ask an assistant instead.
The practical consequence for a local business is that "ranking #3 for roof repair" is becoming a less complete description of visibility than it used to be. You also need to be the business the model names when somebody asks "who should I call about a leaking roof in Lapeer?"
The opportunity is that almost nobody is optimising for this deliberately yet. The techniques are not exotic, and the competition is thin in a way traditional SEO has not been since about 2010.
How generative engines decide who to cite
These systems are not ranking algorithms in the traditional sense. Broadly, two things determine whether you get cited.
Retrieval. When asked a question, the system searches — often using a conventional search index underneath — and pulls in candidate documents. So traditional SEO still matters enormously: if you cannot be found, you cannot be retrieved, and if you cannot be retrieved you cannot be cited.
Synthesis. The model then reads the candidates and writes an answer, choosing which sources to lean on. This is where GEO diverges from SEO, because the model is selecting for things a ranking algorithm never measured: whether a passage directly answers the question, whether claims are specific and verifiable, and whether the source reads as authoritative and internally consistent.
What actually helps
1. Answer the question in the first sentence
The single highest-leverage change. If a page is titled "How much does a website cost?", the first sentence should state a number. Marketing copy that warms up for three paragraphs before reaching the point gives a model nothing extractable, so it uses a competitor who got to the point.
Write each answer so it survives being lifted out of the page entirely. An answer beginning "As we mentioned above..." is useless to a system quoting one paragraph.
2. Be specific and verifiable
"Affordable pricing" and "fast turnaround" are unquotable. "Starting at $4,500" and "four to eight weeks" are quotable. Numbers, dates, named standards, defined processes and explicit constraints all make content more likely to be used, because a model synthesising an answer prefers concrete claims it can attribute.
3. Structured data
Schema.org JSON-LD hands a machine pre-parsed facts instead of making it infer them from
markup. For a local business the high-value types are LocalBusiness (with
accurate areaServed and geo), Service,
FAQPage, BreadcrumbList and Article.
FAQPage deserves particular attention: it hands the engine an explicit
question/answer pair with zero extraction work. That is as close to pre-formatting your
content for citation as the standard allows.
4. Entity clarity
A model needs to know unambiguously who you are, where you are, and what you do. State it plainly in prose, not just in markup: business name, location, services, service area. Consistency matters — if your business name appears three different ways across the web, you have made yourself harder to identify as a single entity.
5. Corroboration elsewhere
Models weight claims that appear in more than one independent place. A fact stated only on your own website is weaker than the same fact appearing on your Google Business Profile, in a trade association directory, in local press and in supplier listings. This is the GEO analogue of link building, and it works for similar reasons.
6. An llms.txt file
An emerging convention (see llmstxt.org) is a plain-Markdown file at
/llms.txt summarising what the site is and linking to its key pages. Support
is not universal and it is not a ranking factor, but it is cheap, it cannot hurt, and it
makes your key facts trivially parseable for any agent that does look.
7. Let the crawlers in
Check your robots.txt. Some sites block GPTBot,
ClaudeBot, PerplexityBot and similar agents by default, often
without the owner realising. Blocking them guarantees you are never cited. Publishers
protecting paid content have a genuine reason to block; a contractor who wants the phone
to ring does not.
Three things that do not work
- Keyword stuffing, in any modern form. Repeating "Lapeer web design" thirty times does nothing for a language model and has done nothing for Google in fifteen years.
- Bulk AI-generated filler. Publishing forty generic articles produces exactly the undifferentiated text a model already generates without you. You get cited for saying something specific that is not already in the model's weights.
- Hidden text or instructions aimed at AI. Attempting to plant directives in hidden markup for a model to follow is treated as spam by search engines and ignored or penalised by serious systems. Do not do it.
How to measure it
GEO measurement is genuinely immature. There is no Search Console for AI citation. What you can do today:
- Ask the engines directly. Query ChatGPT, Perplexity, Claude and Google AI Overviews with the questions your customers ask, and record whether you appear. Repeat monthly.
- Watch referral traffic in GA4 from
chatgpt.com,perplexity.aiand similar sources. It is usually small but growing, and it converts unusually well because the visitor arrives pre-qualified. - Track branded search volume. A rising number of people searching your business name directly is often the downstream effect of being mentioned in answers.
The bottom line
GEO is not a replacement for SEO and anyone selling it as a separate product is overselling. Roughly 70% of the work is excellent traditional SEO — be crawlable, be fast, be authoritative. The remaining 30% is a genuine shift in how you write: direct answers, specific facts, clear entities, and structured data that leaves nothing to infer.
The businesses that will win citations in three years are the ones structuring their content for it now, while almost nobody is.
We build AI-readable structure into every site and offer it as part of our SEO service. This site's own llms.txt is a working example.