GEO optimization — how to get cited by ChatGPT, Perplexity and Google AI Overviews

What GEO is, and why it matters now

An AI assistant does not return ten options. It returns one answer, usually with two to five cited sources. Either your domain is in that short list or, for that question, you do not exist. There is no equivalent of ranking fifth and still collecting traffic.

The scale is no longer speculative. ChatGPT passed 900 million weekly users in early 2026 and has been closing on a billion. Google AI Overviews is live across Europe, though coverage varies sharply by market — a Semrush analysis in 2026 found AI Overviews appearing in up to 17% of European search results while Germany barely broke 1%. Perplexity, Gemini and Copilot add their own audiences on top.

The practical consequence for a business: a share of the research your buyer used to do on Google now happens inside a chat window, and it ends with a recommendation rather than a list. GEO is how you get into that recommendation.

GEO does not replace SEO. ChatGPT with search and Google AI Overviews both draw from what already ranks — if you are nowhere in the top 20 for a query, the assistant will rarely see you. Treat GEO as a layer on a technically sound site, not a substitute for one.

How each assistant picks its sources

The four systems that matter behave differently, and the differences decide where your effort goes. The column that matters most is the last one: how quickly each reflects a change you make today.

Two practical conclusions. First, the fastest wins come from ChatGPT with search and Perplexity, because both read the live web. Second, AI Overviews is downstream of ordinary SEO — the work that gets you there is the work you were already supposed to do.

The one setting that makes you invisible: robots.txt

Before any content work, check which crawlers you allow. A single wrong Disallow is the most common reason a company is never cited, and no amount of content quality compensates for it. Open yourdomain.com/robots.txt and look for these:

If you do not want your content used for model training but do want the citations, the compromise is specific: disallow GPTBot and allow OAI-SearchBot. They are separate crawlers with separate purposes, and most blanket blocks throw away the citations along with the training.

Ten tactics that actually move AI visibility

Ordered by how much they change outcomes relative to the effort they take.

1. Answer the question in the first two sentences

Assistants extract passages, not pages. Put a complete, self-contained answer directly under the heading that asks the question, then expand. A paragraph that only makes sense after three paragraphs of context will not survive extraction.

2. Make every heading a question a real person types

Headings phrased as questions match conversational queries far better than noun phrases. How much does website maintenance cost? beats Maintenance pricing, and the answer underneath becomes directly quotable.

3. Use specific numbers instead of adjectives

“Competitive pricing” is unquotable. “Maintenance packages: €90, €140 and €320 per month” is a sentence an assistant can lift verbatim. Concrete figures, dates and names are the single strongest citation magnet.

4. Add FAQPage and Article structured data

JSON-LD tells machines which text is the question and which is the answer. Pair it with a visible FAQ section — the schema and the page must agree, or you get neither the rich result nor the citation.

5. Build topic clusters, not isolated posts

One in-depth article rarely establishes authority. Eight to ten linked articles covering one topic from different angles, all pointing to a service page, is the structure assistants read as subject expertise.

6. Publish first-hand data nobody else has

Your own numbers — client results, production measurements, pricing, failure rates — cannot be paraphrased from a competitor. Original data is the most reliable way to become the cited source rather than one of twelve similar pages.

7. Name the author and show credentials

A named author with a role, a biography and a verifiable profile, marked up as Schema.org Person, materially changes how a source is weighted. Anonymous content loses ties.

8. Keep a visible last-updated date and honour it

Freshness matters more in AI retrieval than in classic ranking. Review each important page quarterly: refresh figures, remove claims that aged badly, and update the date only when you actually changed something.

9. Write the comparison page your buyer searches for

“X vs Y”, “alternatives to X”, “is X worth it” are queries assistants answer constantly and most companies never write for. A balanced comparison that admits where you are not the right choice gets cited more, not less.

10. Track AI referrals in GA4 from day one

Create a segment for chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Without it you cannot tell whether any of this worked, and most agencies still do not measure it.

A 30-day plan you can actually run

If you do nothing else, do this in this order. Week one is the one that unblocks everything else.

What not to do

Five mistakes that waste a GEO budget, in the order we see them most often.

• Blocking the search crawler by accident. Usually copied from a template robots.txt or added during a “stop AI scraping” panic. Check before you write a single word of content.
• Mass-producing thin articles. Assistants deduplicate aggressively. Twenty generic posts on the same topic produce fewer citations than one article with original data.
• Inventing statistics. Fabricated multipliers get repeated back at you by a client eventually, and an assistant that cross-checks a claim against better sources drops yours. If you cannot name where a number came from, do not publish it.
• Treating GEO as a replacement for SEO. If your site is slow, thin or unindexed, GEO work has nothing to stand on. Fix the foundation first.
• Measuring nothing and declaring success. Without the GA4 segment and a repeated citation check, “we did GEO” is an invoice, not a result.

How to measure whether it is working

Three measurements, all free, all repeatable monthly:

• AI referral sessions in GA4. A segment covering chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Watch the trend, not the absolute number — volumes are small but the visitors convert unusually well, because they arrive already advised.
• A manual citation check. The same ten buyer questions, asked monthly in ChatGPT with search, Perplexity and Gemini. Record whether you are cited and who is cited instead. This is the only direct measurement of GEO that exists today.
• Search Console question queries. Filter for queries containing question words. Growth there means your content is matching conversational intent, which is the same intent assistants serve.

Expect first citations within one to two weeks for pages that are already indexed, and four to eight weeks for genuinely new content. Google AI Overviews is slower — two to three months is realistic, because it inherits classic ranking lag. These are ranges from practice, not guarantees; domain authority and niche competition move them in both directions.

Frequently asked questions

Key takeaways

• AI answers are binary: you are cited or you are invisible for that question.
• Check robots.txt first — OAI-SearchBot blocked means nothing else you do matters.
• Write answers, not pages: one question per heading, the answer in the first two sentences.
• Specific numbers and first-hand data get cited; adjectives do not.
• GEO sits on top of SEO, it does not replace it.
• Measure with a GA4 AI-referral segment and a repeated manual citation check, monthly.

Last updated 12 November 2026. Figures on assistant usage and AI Overviews coverage reflect reporting available at that date and are summarised, not quoted; verify current numbers before reusing them.