How to Show Up in ChatGPT and Google AI Overviews: A 2026 GEO Playbook

Last updated: August 2026 · 12 min read

What GEO is, and what changed in 2026

Generative engine optimization is the work of getting your brand named, cited, and recommended inside AI generated answers. The distinction from classic SEO is not cosmetic. In search, you optimize a page to win a position, and the user still chooses which link to click. In an AI answer there is no list to choose from. The model has already chosen, and it usually names between three and fifteen sources. Everyone else is invisible.

That compresses the funnel in a way most teams underestimate. Position 8 on Google still earns clicks. There is no position 8 in an AI answer. This is why GEO rewards being the clearest, most quotable, most corroborated source on a narrow topic rather than the best all round page.

The other change is that the two disciplines have stopped being separable in practice. Semrush's 2026 AI Visibility Index reported that 81% of organizations running SEO and AI visibility as one integrated strategy saw increased traffic or leads from AI platforms, against 36% of those managing them separately. If you are running GEO as a side project owned by someone else, that gap is the cost.

How AI engines actually choose sources

Every major engine now works in two stages: retrieve a candidate set from the live web or an index, then select a small number of sources to ground and cite the answer. Optimising only for stage one is the most common mistake, because stage one is where SEO habits transfer and stage two is where they do not.

The selection step is far narrower than search results, and it is not consistent between engines. Semrush's index, which analyzed 126 million United States AI search prompts between January and April 2026 across ChatGPT, Gemini, Google AI Mode and AI Overviews, found that ChatGPT cites an average of 15 sources per response while Gemini cites around 3. The practical consequence: Gemini is close to winner takes all, and ChatGPT gives a mid tier brand a genuine route in.

The same study found only 36 brands maintained visibility across every platform, a group including YouTube, Google, Reddit, Amazon, Facebook, Apple, Walmart, Disney and Nintendo. Read that as permission rather than discouragement. If near enough nobody is universally visible, per engine gaps are normal, and the win is picking the engine where you are closest to breaking through instead of chasing all of them at once.

Both ChatGPT and Gemini lean heavily on Wikipedia and Reddit. That is the single most actionable finding in the whole dataset, and step four is built around it.

Step 1: Baseline audit

You cannot improve what you have not measured, and almost nobody has measured this. Give yourself an afternoon.

Run the free check first. Semrush's AI Search Visibility Checker returns a baseline for your domain without a subscription, and it is the fastest way to see whether you appear at all. Treat the output as a starting point rather than a verdict.

Then check by hand. Pick ten prompts a real buyer would type, open each engine in a logged out or incognito window so personalisation does not distort the result, and record three things for each: whether you are mentioned, which competitors are, and which domains get cited as sources. Log it in a spreadsheet with the date and the engine. That last column matters more than the other two, because the cited domains are your citation building target list.

Audit whether the machines can read you. Check that AI crawlers are not blocked in robots.txt, that key pages render their content in raw HTML rather than only after JavaScript executes, and that your pages are reachable within a few clicks. A page an AI crawler cannot fetch cannot be cited, and this is the failure mode we see most often on otherwise healthy sites. Semrush's AI Search Site Audit automates the check; a manual curl of your page and a look at what actually comes back does the same job for free.

Step 2: Build the prompt set

A prompt set is the GEO equivalent of a keyword list, and it is where most programs go wrong, because people transliterate keywords into prompts. "best crm software" is a keyword. "I run a 12 person agency and need a CRM that syncs with Gmail, what should I use" is a prompt. The second is what people actually type, and it is the one that produces a recommendation.

Build across four intent bands so you can see where you break down:

• Category prompts. "What is the best AI writing tool" and its variants. Highest volume, hardest to win, usually dominated by incumbents.
• Qualified prompts. The category plus a constraint: budget, company size, integration, industry. This is where a mid tier brand realistically wins.
• Comparison prompts. "X versus Y", "alternatives to X". Often the highest intent of the four, and the easiest to influence with a genuinely useful comparison page.
• Brand prompts. "Is X any good", "is X worth the money". You will not always like the answer, and this is exactly why sentiment tracking matters.

Twenty five well chosen prompts beat two hundred lazy ones. Semrush's Prompt Research report gives topic level volume, difficulty and intent, which is the closest thing the category has to keyword research for prompts, and it is the fastest route to a defensible list. If you are building manually, mine your sales team's call notes: the questions prospects ask on a first call are almost verbatim the prompts they typed the week before.

Step 3: Content and entity optimization

Now make your content easy to lift. Models quote passages that stand on their own, so the writing changes are concrete and they are not subtle.

Write extractable answers

Lead each section with a direct, self contained answer of two or three sentences, then support it. If a paragraph only makes sense after reading the two above it, a model cannot quote it. Put the definition immediately under the heading that asks for it. Use headings that mirror real questions rather than clever labels.

Be specific and be checkable

Numbers, dates, named prices, named versions, and stated limitations all raise the odds of citation, because they are the parts of a page that answer a question precisely. Vague marketing copy is unquotable by construction. Where a fact will age, date it on the page so a model can tell how fresh it is.

Consolidate the entity

An entity is your brand plus everything the model believes about it. Models build that picture from many sources agreeing, so contradictions are expensive: if your pricing page, your G2 profile, and three roundups all state different numbers, the model has no consensus to draw on and tends to route around you. Pick the canonical facts about your company and make them identical everywhere you control, then chase down the places you do not.

Keep the structured data honest

Organization, Product, and FAQPage schema help machines parse what a page asserts. Mark up what is genuinely on the page and nothing else. Fabricated review markup is both a policy violation and, increasingly, a way to teach a model that your domain is unreliable.

Step 4: Citation building

This is the step with the clearest evidence behind it and the one most teams skip, because it happens off your own site.

Since ChatGPT and Gemini both lean on Wikipedia and Reddit, being absent from both is a structural disadvantage no amount of on site work fixes. The honest version of this work looks like:

• Wikipedia, indirectly. Do not write your own entry and do not pay anyone who offers to. Notability is established by independent coverage, so the real task is earning the sort of sourcing an editor could legitimately cite. If your category has a page, making sure the facts on it are accurate and sourced is fair game.
• Reddit, by being useful. Find the subreddits where your category is genuinely discussed and take part as yourself, disclosing who you work for. Astroturfing gets detected, gets punished, and produces exactly the negative sentiment the tracking in step five will surface.
• The domains your baseline surfaced. Every cited domain from step one is a target: review sites, industry publications, comparison pages, listicles. Getting accurately represented there compounds, because those pages are already in the retrieval set.
• Third party review platforms. G2, Capterra and their sector equivalents are heavily retrieved for software questions, and they are one of the few places where volume and recency of real customer reviews directly moves what a model says about you.

The through line: AI answers reward consensus across independent sources. One excellent page you own is worth less than five credible pages you do not.

Step 5: The measurement loop

AI answers are non deterministic. The same prompt on the same engine can return different sources an hour apart, so a single check tells you almost nothing and reacting to one is worse than not looking. You need repeated sampling and a trend.

Track four things and no more:

• Visibility. How often you appear across your prompt set. Semrush expresses this as an AI Visibility Score benchmarked from 0 to 100 against competitors, which is the number to put in front of a board.
• Share of voice. Your mentions as a percentage of the competitive set. This is the one that reveals whether you are gaining or the whole category is.
• Citations. How often your domain is used as a source, as opposed to your brand simply being named. These move independently and the gap between them is diagnostic.
• Sentiment. How you are characterized. Being mentioned as the expensive option is not the same win as being mentioned as the best one, and only sentiment tracking tells them apart.

Sample weekly, judge monthly, and change one variable at a time. The loop closes when a content or citation change shows up as a sustained move in share of voice rather than a one week spike.

Worked examples on real domains

Rather than invent case studies, here are three real, checkable situations and what the playbook says to do about each. You can verify every number cited here against the source study.

Reddit and Wikipedia: the two domains everyone competes against

These are not competitors in your category, but in retrieval terms they are the domains you are most often losing a citation slot to, on both ChatGPT and Gemini. You cannot outrank them and should not try. What you can do is be the source they cite, or be accurately represented on them. For a software brand that means a factually correct presence in the relevant subreddit discussions and, if you are notable enough to have an entry, an accurate and well sourced one. This is step four, and it is the highest leverage work available to most brands.

A News and Media brand: 82.9% of visibility goes to three players

If you publish in News and Media, the index found the top three brands take 82.9% of category visibility. Competing for "what is happening with X" against that concentration is not a realistic near term goal. The playbook response is to abandon the head entirely and own qualified and comparison prompts where the incumbents are thin: specific sub topics, specific regions, specific recurring questions. Measure share of voice inside that narrow set rather than category wide, or the number will never move and you will conclude wrongly that GEO does not work.

An Industrial or Finance brand: 41.4% to 42.2% concentration

These categories are wide open by comparison, and the strategy inverts. Head prompts are winnable, so go after them directly with genuinely authoritative content and aggressive entity consolidation. The window here is a function of competitors not yet doing this work, which means the cost of waiting is higher in an open category than a closed one, not lower.

One thing we will not do is publish a screenshot of a live ChatGPT or Gemini answer as evidence. Those outputs vary by session, account, and region, so they are not reproducible, and a result we captured today would not be what you see when you run it. Run the prompts yourself as described in step one; that baseline is the only one that is true for you.

The monthly workflow

Manual, suite, or dedicated tracker

There are three honest ways to run the measurement layer, and the right one depends on the size of your prompt set.

We use Semrush's toolkit as the measurement layer in this playbook because it is the one that reports across the widest set of engines and connects the finding to the SEO data you need to fix it. Its Brand Performance and AI visibility reports cover Google AI Overviews, Google AI Mode, ChatGPT, Perplexity and Gemini. Worth knowing before you buy: Prompt Tracking, the daily prompt level view, currently runs on ChatGPT, Google AI Mode and Gemini, with Perplexity and AI Overviews appearing in the brand level reporting instead. The toolkit is sold standalone from $99/mo billed annually, covering 25 tracked prompts and one domain, with 50 extra prompts at $60/mo. There is no free trial on the toolkit itself, so start with the free checker.

If you would rather compare the whole category first, our guide to the best AI search visibility tools of 2026 scores nine of them on a published framework, and how to track AI citations goes deeper on the measurement mechanics. For the traditional half of the work, see the best AI SEO tools of 2026.

Frequently asked questions

Disclosure: this article is a sponsored partnership with Semrush and contains affiliate links. Our editorial recommendations, including the independent alternatives named above, are made on merit and are not changed by the partnership. Statistics are from Semrush's 2026 AI Visibility Index (126 million United States prompts, January to April 2026). Semrush pricing and feature coverage verified 4 August 2026; confirm current figures on the vendor's site.