My SEO Is Dialed In. Where Do I Start With GEO?

You have already done most of it

Here is the thing nobody selling GEO services will lead with: if your technical SEO is genuinely solid, you have already done perhaps 80 percent of the work. AI systems retrieve from search indexes, so the prerequisites are the same ones you already met.

A page that is crawlable, indexed, fast, well-structured and authoritative is a page an AI system can retrieve and quote. There is no separate index to get into.

Which means the honest starting move is not a project. It is finding out where you already stand, because most well-optimised sites appear in AI answers more than their owners expect.

The four things that are genuinely new

Strip out what overlaps with SEO and this is what is left:

• AI crawler access is a separate decision. GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot and Google-Extended are governed from your robots.txt, and many sites block them without ever having decided to.
• Measurement does not exist in Search Console. There is no report for "how often were we cited", so you assemble it manually.
• The unit of success changed. Not a position, but a mention, citation or recommendation, measured as prompt coverage across a fixed set of questions, against competitors.
• Quotability is now a property worth engineering. A claim that can be lifted cleanly gets cited more than one that has to be paraphrased.

Week one: find out where you stand

Three checks, in this order, because the first two make the third meaningful:

1. Check AI crawler access. Run your domain through our robots.txt tester, which lists AI crawlers explicitly. If the search crawlers are blocked, stop here and fix that first, because nothing else matters until you are retrievable.
2. Confirm your content renders without JavaScript. Some AI crawlers render JS and some do not. If your key content only appears after hydration, you are invisible to the ones that do not.
3. Run a manual prompt test and write down the result. Twenty buyer-intent questions across ChatGPT, Claude, Perplexity and Gemini. Log three outcomes separately: mentioned, cited with a link, or actively recommended. Note who appears when you do not.

For each of these 20 questions, record per assistant: (1) were we mentioned by name, (2) were we cited with a link, (3) were we recommended over alternatives, (4) which competitors appeared. Same 20 questions, same wording, monthly. Do not vary the questions, because the whole value is comparability over time.

That log is your baseline and it is free. Most tools in this category are automating exactly this, less reliably, because the surface is genuinely unstable between runs.

Week two: the structural changes

The content changes that improve citation are, usefully, changes that also improve the page for humans:

• Answer the question in the first 50 words. A page that buries its answer under context is harder to quote cleanly, and harder to read.
• Make claims specific and self-contained. "Audits typically cost $2,000 to $7,500 for a mid-size site" can be lifted. "Costs vary considerably" cannot.
• Keep entity signals consistent. The same company name, the same product names, the same descriptions across your site and your off-site profiles.
• Add the data only you have. Original figures get cited because there is nowhere else to get them, which is the most durable advantage available here.

None of that trades against ranking, which is what makes this transition easier than previous ones. The full tactic list is in our LLM SEO guide.

Adding GEO to an existing retainer

The practical question if you are doing this for clients. Two line items cover it honestly:

1. A monthly AI visibility report. The prompt test above, run against a fixed question set, with competitor appearance tracked. This is the deliverable clients actually want, because it answers "are we in there".
2. A brief-level change to content work. Answer-first structure and quotable claims become part of every brief rather than a separate service.

Making the measurement survive contact with reality

The prompt test is easy to start and easy to abandon, because the results are noisy enough to feel meaningless in month two. Three habits keep it useful:

• Never change the questions. The value is comparability, and a rewritten question set resets your baseline to zero without telling you.
• Record the competitor, not just your absence. "We were not mentioned" is a data point. "We were not mentioned and the same three competitors were" is a brief.
• Expect run-to-run variance. The same question can produce different sources an hour apart. Read the trend across months, and treat a single bad run as noise rather than a regression.

That last point is why I am cautious about tools promising precise AI visibility percentages. The surface genuinely varies by user, session and time, so a number reported to one decimal place is describing sampling noise with unearned confidence.

Which of your pages to prioritise

Not all of them, and the selection is not the same as your SEO priority list. The pages worth structuring for citation are the ones where being the quoted source has commercial value:

• Comparison and evaluation pages. Someone asking an assistant which tool to choose is close to a decision, and a citation there is worth more than ten on a definitional query.
• Pages carrying original data. Your own benchmarks, survey results or test findings get cited because there is nowhere else to source them.
• Anything answering a question with a specific number. Prices, ranges, timelines and thresholds are exactly what a generated answer needs and what vague content cannot supply.

Pages that summarise general knowledge are the least worth optimising for citation, because a model does not need a source for something it can state directly. That is the same judgement as the intent-mix argument: the exposure and the opportunity both concentrate in the same place.

What not to bother with

• Rebuilding content that already ranks. If it ranks, it is retrievable. Add structure where it is cheap; do not rewrite what works.
• Buying an AI visibility tool before establishing a baseline. Do the manual test first. You may find you are doing fine, which changes what the tool is worth.
• Chasing every new acronym. GEO, AEO, LLM SEO and AI SEO substantially describe the same work. Pick one label and move on.
• Blocking AI training crawlers as a strategy. That is a licensing decision, and a legitimate one, but it does not improve your visibility. Just do not block the search crawlers by accident while doing it.