Short answer
Generative engine optimization (GEO) is the work of making your pages easy for AI answer engines to find, read and cite. Think of ChatGPT search, Perplexity, Claude, Gemini and Google’s AI Overviews. It rests on ordinary SEO: crawlable, indexable pages with clear text. What changes is the goal, being quoted or linked inside an answer, and how you measure it. Analytics cannot see the answer itself, but it can show two things around it: which AI crawlers read your pages, and how many people click through from an assistant.
Where does the term GEO come from?
The name comes from a research paper, GEO: Generative Engine Optimization, by researchers at Princeton University, IIT Delhi and independent co-authors. It was first posted in November 2023 and published at the ACM KDD conference in August 2024. The authors define a generative engine as a search system that gathers several sources and writes one answer with an LLM, and they describe the problem for site owners plainly: you have little control over whether and how your content shows up in that answer.
They built a benchmark of 10,000 queries (GEO-bench) and tested nine ways of rewriting a source page. The headline result is that the best methods raised a source’s visibility in generated answers by up to 40%. The detail matters more than the headline:
- Adding quotations, statistics and citations to sources performed best across their metrics.
- Keyword stuffing, the classic SEO trick, did little or nothing compared with the unchanged page.
- Easier, more fluent text also helped, by roughly 15–30% in their tests.
- Results varied by domain. A method that worked for history questions did not always work for law or health.
Keep the setup in mind: the experiments used GPT-3.5 Turbo as the generative engine plus a smaller test on Perplexity. Today’s assistants use other models and retrieval steps, so read the paper as a direction (be specific, cite, show numbers), not as a recipe with fixed percentages.
How is GEO different from SEO?
Less than many GEO vendors suggest. Google’s own page on AI features and your website says the SEO best practices still apply to AI Overviews and AI Mode, that there are no additional technical requirements, and that you don’t need new machine-readable files, AI text files or special schema.org markup. A page has to be indexed and eligible for a snippet to be shown as a supporting link. Google also notes that meeting the requirements does not mean a page will be crawled, indexed or shown.
The differences sit in three places:
- The unit of success. In classic search you win a position and a click. In an AI answer you win a mention, a quote or a source link, and the visitor may never click.
- The way questions are asked. People type long, full-sentence questions into assistants. Pew Research found that Google searches of 10 words or more produced an AI summary 53% of the time, against 8% for one- or two-word searches.
- More crawlers, with different jobs. Besides Googlebot and Bingbot, your pages are now fetched by training crawlers, AI search crawlers and on-demand fetchers. Each one is a separate robots.txt decision.
Do AI answers still send clicks?
Fewer per search, and that is why measurement matters. In Pew Research Center’s analysis of the browsing data of 900 U.S. adults in March 2025, people who saw a Google AI summary clicked a traditional result in 8% of visits, against 15% when there was no summary. They clicked a link inside the summary itself in just 1% of visits. Around 18% of the Google searches in that data produced an AI summary.
Google’s view is that clicks from pages with AI Overviews are of higher quality, with visitors spending more time on the site. Both can be true: fewer clicks, but more intent behind each one. The only way to know which applies to your site is to count those visits separately and see what they do next.
What should you measure for GEO?
Most GEO advice stops at “track your AI visibility”. Split it into what happens before, during and after an answer, because each step needs a different source of data.
Before the answer: are AI crawlers reading your pages?
An assistant can only cite what its crawler fetched or what a search index gave it. OpenAI documents OAI-SearchBot for ChatGPT search, GPTBot for training and ChatGPT-User for fetches a person triggers. Anthropic and Perplexity publish similar splits. Track requests per bot and per page, so you can answer: does the search crawler ever read my pricing page? Did the bot I blocked actually stop? These requests never run JavaScript, so a browser tag such as GA4 does not see them. You need server, CDN or edge logs. Our guide on how to track GPTBot and other AI crawlers covers the setup.
During the answer: are you mentioned or cited?
This part is invisible to any analytics tool, including ours. The answer is generated inside the assistant. To see it you either run a fixed list of prompts in each assistant on a schedule and log whether your brand or URL appears, or you use a prompt-tracking tool that does this for you. Either way, answers vary by user, location and day, so treat the result as a sample, not a ranking.
After the answer: do people click through, and do they buy?
This is where analytics is strongest. Count visits whose referrer is chatgpt.com, perplexity.ai, claude.ai, gemini.google.com or copilot.microsoft.com, and visits tagged with parameters such as utm_source=chatgpt.com. Then look at their landing pages, goals and revenue next to your other channels. Some assistant clicks arrive without a referrer and fall into direct, so this number is a floor, not the full picture. Our post on how to see ChatGPT traffic in analytics shows the GA4 and non-GA4 ways to do this.
Which GEO tactics are worth doing?
Start with the ones that also help human readers and classic search, and that the research supports:
- Answer first. Put a direct two- to four-sentence answer at the top of the page, then the detail. Assistants quote short, self-contained passages.
- Use question headings. Write H2s the way people ask, for example “How is GEO different from SEO?”, so a passage matches a long question.
- Add numbers and sources. Statistics, quotations and cited sources were the strongest methods in the GEO paper. Link the source and say when you checked it.
- Keep important content in text. Prices, specs and comparisons hidden in images, tabs that load late or PDFs are harder to read for every crawler.
- Make structured data match the page. Article, FAQ and Product markup is fine, as long as it says what the visible text says. Google is explicit that no special AI schema exists.
- Decide per crawler in robots.txt. Allowing
OAI-SearchBotwhile disallowingGPTBotis one split used by sites that want to be cited but not used for training. OpenAI says robots.txt changes for search take about 24 hours to apply.
Be careful with anything sold as a shortcut: hidden prompt text, mass-generated pages or claims that a tool can place you in answers. Google’s spam policies already cover hidden text and scaled, mass-produced content, and no assistant publishes a ranking you could buy into.
How does Wholetally help you measure GEO?
Wholetally covers the before and after steps. It does not track mentions inside answers.
- The AI report shows AI crawler requests per bot and per page, split into training, AI search and user request, next to Googlebot and Bingbot for comparison. Crawlers are recognised by user agent at the edge; bots that are not on our list show up as unknown bots, not as people.
- The same report counts visitors referred by AI assistants as one channel, with their landing pages and goals, including utm tags like
utm_source=chatgpt.com. - Edge collection is what makes crawler counts possible, because those requests never run a script in the browser.
- Revenue per source links Stripe payments back to the channel, so you can see whether assistant visitors buy, not only whether they arrive.
The script is cookieless and creates no visitor ID; the what we collect page lists every field. To see how this compares with GA4 and other privacy-friendly tools, use the comparison table.
What to do next
Pick five pages you want assistants to cite, usually pricing, your main product or service pages and your best guide. Check that AI search crawlers can fetch them, rewrite the top of each page as a direct answer with a number and a source, and then watch two numbers for a month: AI search crawler requests to those pages, and visits referred by assistants. If both move, your GEO work is reaching people. Try Wholetally next to your current setup with a 30-day free trial, no card needed. See plans on Pricing.
Wholetally is cookieless web analytics with an AI report, edge collection and revenue per source. Built in the Netherlands. Research figures and vendor documentation were checked on 15 November 2026 and may change.