Generative engine optimization (GEO): how AI engines cite your brand

Short answer: generative engine optimization (GEO) is the practice of preparing your content so that engines like ChatGPT, Perplexity, Google AI Overviews and Claude cite it as a source while they compose an answer. Classic SEO fights for a position among ten blue links. GEO fights to be named inside the answer itself. Three things decide it: HTML a machine can actually read, paragraphs that survive being cut out of context and carry hard numbers, and a brand entity described consistently everywhere it appears.

What is generative engine optimization (GEO) and how does it differ from SEO?

Generative engine optimization is the work of preparing a page so a model borrows a sentence, a number or a definition from it while writing an answer, rather than so the page climbs a list. Classic SEO is a race for position: there are ten results, and you are somewhere in them. GEO is a selection process: the engine decides which three to five sources it will lean on, then paraphrases them. Traditional signals (technical health, authority, links) still matter, because they make the engine trust you. They are simply no longer enough. If the model cannot find one clean, quotable sentence on your page, it moves to the site that has one.

• The goal changes: not ranking, but being cited. Position one does not guarantee a mention in the AI answer.
• The unit changes: not the page, the paragraph. The engine lifts a single block that makes sense on its own.
• Click behaviour changes: industry compilations report a sharp fall in organic click-through on queries where an AI Overview appears, while the smaller stream of visitors arriving from AI answers tends to be better qualified.
• The competition changes: forums, listings, review pages and technical documentation are cited far more often than polished corporate blogs.
• Measurement changes: instead of impressions and average position, you track citation frequency, AI referral domains and branded search lift.

What kind of content do AI engines actually cite?

AI engines cite content that is verifiable and easy to lift. Six criteria do most of the work: fact density (numbers, thresholds, formulas, dates), a plain definition of the key term at first mention, self-contained paragraphs of roughly 40 to 75 words that still make sense out of context, structured data (JSON-LD for author, dates, questions and answers), freshness with a visible update date, and crawlable HTML that does not need JavaScript to exist.

• Engines pull from very different places. Perplexity puts sourcing at the centre of its architecture and lists more than 20 links per answer, while ChatGPT stays below 10. In 2026 compilations, only around 10% of the domains cited by ChatGPT are also cited by Perplexity: you cannot optimise for one engine and assume the rest will follow.
• In Yext's analysis of 6.8 million AI citations, about 86% of cited sources were surfaces the brand manages: close to half the brand's own website, the rest listings, reviews and structured directories. Analyses built on a different method give third-party sources a larger share, so do not treat the number as absolute; the weight of your own site comes out high in every measurement.
• The GEO study from Princeton and IIT Delhi, published at the KDD 2024 conference, tested nine content tactics across 10,000 queries: adding statistics, quotations and explicit sources lifted visibility by up to 40%. In the same study, keyword stuffing pushed visibility down.
• AI Overviews are no longer an exception. Method changes the answer: depending on the keyword set and the country, 2026 trackers report AI Overviews on somewhere between a quarter and a half of Google searches. The spread is wide, the direction is not.
• Traffic from AI engines is still small on most sites: 2026 compilations put the average at roughly 1% of total sessions, and noticeably higher in verticals such as B2B technology. It converts above classic organic search, though, with different analyses reporting anything from a 40% edge to several times better.

Do AI crawlers run JavaScript?

No, they do not. This is the most skipped and most expensive item on the list. Crawlers such as GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot fetch the raw HTML, read whatever is inside it and move on: they do not execute JavaScript, they do not wait for a render, they do not come back for a second attempt. In the Vercel and Merj study of crawler logs, more than 500 million GPTBot requests produced no evidence of JavaScript execution at all. The one meaningful exception sits on Google's side, where Gemini can lean on Googlebot's rendering infrastructure. The consequence is uncomfortable: a client-rendered site can rank perfectly well in Google and still be a blank page to ChatGPT or Perplexity.

1. Server-side rendering: ship the content in the first HTML response. Do not leave text, headings and links to JavaScript. The test is simple: fetch the page with curl. If you cannot see the text, neither can the engine.
2. robots.txt decisions: decide deliberately about GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended. If you want to be cited, do not block them.
3. llms.txt: publish a plain-text map at the root of your domain that lists your most important pages with a one-line description of each.
4. Schema markup: add JSON-LD for Article/BlogPosting, FAQPage, Organization and BreadcrumbList. Make author, publish date and update date visible.
5. Question-shaped headings: turn H2s into real questions and put the self-contained answer immediately underneath.
6. Entity consistency: write your brand name, description and category identically on your site, in listings and on profiles. Models recognise an entity through that repetition.

How do you measure AI engine traffic in GA4?

You measure AI traffic by isolating referral domains. In May 2026 GA4 added an 'AI Assistant' channel to its default channel group, so sessions from recognised AI assistants now sit on their own line. The coverage is incomplete, though. Google has not published the full list of recognised referrers, industry tests still find surfaces such as Perplexity landing under Referral, clicks from AI Overviews count as Organic Search, and the new classification is not applied to historical data. Above all: a large share of the sessions that start inside an AI answer arrive with no referrer and land in Direct. So build your own channel group, read branded search and direct traffic beside it, and add the section to your client reports as a standing block.

• Build your own channel group: the built-in 'AI Assistant' channel in GA4 does not catch every surface. Define an AI assistant channel yourself, matching referral domains such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com.
• Do not wait for UTMs: you cannot tag a link that a model writes. Referral domain plus landing page is your primary evidence.
• Watch branded search: the impression and click trend for brand queries in Search Console is the most reliable indirect proxy for AI visibility.
• Read Direct separately: sessions that lose their referrer pile up there, so expect Direct to rise as your AI visibility rises.
• Run a manual citation check: once a month, ask your target questions in ChatGPT, Perplexity and Google AI Overviews by hand and record who gets cited. Even a simple sheet reveals the trend.

What does AI visibility do for your ads?

Being cited by an AI engine pays back on the paid side in two ways. Volume first: when someone asks ChatGPT which brands are worth a look in your category and sees your name, the next move is almost always a branded search on Google. That widens the impression pool for your brand campaigns, and branded keywords are nearly always the cheapest, highest-ROAS line in the account. Quality second: a visitor arriving from an AI answer has already asked the question and already read the comparison, and industry compilations measure this traffic converting above classic organic. GEO looks like a content project, but it lands squarely on media efficiency. The logic behind AI-powered ad management is identical: a machine can only use the data it is able to read.

AI engines do not rank you, they choose you. To be chosen, you need one machine-readable sentence that stands on its own and carries a real number.The Ads Sensor Team

Where to start: a 90-day GEO plan

1. Days 0 to 15, audit: pull your top 20 pages with curl and confirm the text really is in the HTML. Treat it like an ad account audit: find the leaks before you spend on anything new.
2. Days 15 to 30, question map: list the 30 to 50 questions a buyer would put to an AI engine. The ones your sales team hears most are the best starting point.
3. Days 30 to 60, rewrite: on your first ten pages, turn H2s into questions, add 40 to 75 word self-contained answers underneath, and put at least one concrete number in every answer.
4. Days 60 to 75, plumbing: ship JSON-LD (Article, FAQPage, Organization), llms.txt and your robots.txt decisions, and make update dates visible.
5. Days 75 to 90, measurement: build the GA4 channel group, schedule the monthly manual citation check, and record branded search volume as your baseline.

GEO is not a campaign, it is a maintenance discipline: you stay citable for as long as your content stays current, your HTML stays readable and your brand entity stays consistent. The same discipline applies on the paid side, where you cannot see which channel truly works until the data sits in one place. Ads Sensor brings Meta Ads, Google Ads, TikTok Ads, Criteo and GA4 into a single panel, analyses them with AI and produces prioritised, reasoned actions. Create your pre-beta account and watch branded search lift next to the ad results it feeds.