How to Do GEO: Five Steps to Get AI to Recommend Your Brand
GEO (Generative Engine Optimization) is the practice of getting ChatGPT, Gemini, Perplexity and Google AI Overviews to mention your brand and cite your pages. Starting from where AI answers come from, here is how to do GEO in five steps: measure, pick topics, write, make your site readable to AI, and track.
Also available in：繁體中文
GEO (Generative Engine Optimization) is the practice of getting AI tools such as ChatGPT, Gemini, Perplexity and Google AI Overviews to mention your brand and cite your pages when they answer questions. How do you do GEO? It is a five-step loop: measure what AI says about you today, pick topics AI still answers poorly, write content that is easy to cite, make sure AI can read your site, and measure again after about 4 weeks to see whether it worked.
GEO is not a new technology. It is a set of content and website habits that make your pages easy for AI to find, understand and cite. This guide first explains where AI answers come from, then walks through the five steps.
Why does GEO matter now?
People are shifting from clicking links to reading the AI answer directly. When the answer is written inside the AI response, a brand that is not mentioned effectively does not exist.
- When a Google search shows an AI summary, users click a traditional search result 8% of the time, versus 15% without one; only 1% click a link inside the summary (Pew Research Center, July 2025).
- Google AI Overviews has more than 2 billion monthly users (Alphabet Q2 2025 earnings).
- ChatGPT has 800 million weekly active users (TechCrunch, October 2025).
The term GEO comes from a 2023 research paper (Aggarwal et al., GEO: Generative Engine Optimization, later published at KDD 2024). For how GEO differs from SEO and AEO, see GEO, AEO and SEO: definitions and differences.
Where do AI answers come from?
AI answers draw on two sources: the training data the model learned from, and the search results it retrieves at the moment it answers. GEO mainly influences the search results.
| Source | What it is | How fast it updates | Does the answer link to sources? |
|---|---|---|---|
| Training data | Large amounts of text the model read when it was built, stored in the model | Slowly, with the next model version | No |
| Retrieved results | Pages the AI searches for and reads before answering | Quickly, as soon as a page is indexed | Yes, usually next to sentences or below the answer |
The major AI tools search the web before answering questions that need current information. ChatGPT search indexes pages with its own crawler, OAI-SearchBot (OpenAI crawler documentation), and Google AI Overviews cites pages Google has already indexed, with no additional technical requirements (Google Search Central).
Training data is hard to change quickly; retrieved results depend on what is on the web right now. So the core of GEO is making your pages findable when AI searches, and worth citing once found.
How does AI choose which pages to cite?
AI does not cite pages in ranking order. It picks the passages that best answer the question from the pages it retrieved. To be cited, a page must first be found, then be chosen.
The GEO paper rewrote the same set of pages in different ways, had AI answer queries, and measured how much of each answer came from each page:
| Method | What changed | Overall score |
|---|---|---|
| Unoptimized | The original page | 19.5 |
| Keyword stuffing | Repeated search keywords | 17.8 |
| Cite sources | Added verifiable sources | 24.9 |
| Statistics addition | Replaced vague claims with specific numbers | 25.9 |
| Quotation addition | Quoted credible people or organizations | 27.8 |
Citing sources, adding statistics and adding quotations improved visibility by roughly 30–40% relative to the unoptimized page; keyword stuffing made it worse. With sources added, a site ranked 5th in search results gained 115.1%. AI cares about how citable the content is, not only how big the site is.
The five steps of GEO
GEO is a repeating loop: measure, find gaps, fill them, make your site readable to AI, and measure again. Without measurement, you cannot tell whether a change worked.
| Step | Question to answer | Output |
|---|---|---|
| Step 1 Measure | Does AI mention me now? Whose pages does it cite? | A list of which questions you appear in and which you do not |
| Step 2 Pick topics | Which questions does AI answer poorly that I can answer best? | One priority topic |
| Step 3 Write | How do I write so my content gets chosen? | One easy-to-cite article |
| Step 4 Make it readable | Can AI crawlers get in? Can they understand the page? | A site that is open to AI and clearly structured |
| Step 5 Track and adjust | After publishing, did the AI answers change? | What to change next round |
Step 4 is a site-level setup you usually do once; the other four steps repeat for every topic. The first time, check Step 4 first, or even great content will not be read.
Step 1: Measure what AI says about you today
Measuring means asking several AI tools a fixed set of questions on a regular schedule and recording how they answer. The questions should be ones your customers actually ask, and they must not include your brand name, or you cannot measure your visibility when nobody prompts for you.
AI answers vary slightly each time, so ask each question at least 3 times and count how many of the 3 answers mention you. For each answer, record five things: whether you are mentioned, where, whether it is a recommendation or a passing mention, the tone, and whether your site is among the cited sources. Pay most attention to questions where nobody is mentioned, and questions where you are mentioned but someone else's page is cited.
Step 2: Pick topics AI still answers poorly
The best topics are questions your customers often ask, where AI answers are still weak, and where you know the most. Two ways to judge:
- Check scarcity: ask several search-enabled AI tools about the topic and count how many different websites they cite. Topics with almost no sources offer the biggest opportunity.
- Find content gaps: note what AI leaves out, answers vaguely or gets wrong, such as specific costs, step order, or regional and regulatory differences. These are what your article should fill.
Once you pick a topic, add its question to your Step 1 list so you can see the change after publishing.
Step 3: Write content AI can easily cite
When AI picks passages, it favors content where the answer is obvious and backed by evidence. Seven practices:
- Answer directly in the first 100 words, without filler such as "In this article we will"
- Use specific numbers with clear units and conditions
- Cite at least 3 external sources, at least 1 from a government, academic or official body
- Use tables for comparisons and ordered step lists for processes
- End with an FAQ where every question has a one-sentence direct answer
- Show a real last-updated date and author
- Do not stuff keywords
For details and examples of each practice, see AI citation optimization: 5 practical tips.
Step 4: Make your site readable to AI
Great content that AI cannot read might as well not exist. This step checks the site itself and only needs to be done once:
- Do not block AI crawlers: check that robots.txt allows OAI-SearchBot, Googlebot, PerplexityBot, Claude-SearchBot and similar crawlers. OpenAI controls its training crawler GPTBot separately from its search crawler OAI-SearchBot, so you can allow search only.
- Put the text in the page itself: search for a sentence from your article in the browser's "view source". If it is not there, the content only appears after scripts run, and some crawlers will miss it.
- Add structured data: mark up Article (author, dates) and FAQPage (questions and answers) in Schema.org format, matching what is visible on the page.
- Sitemap and llms.txt: a sitemap keeps crawlers from missing new articles; llms.txt is a site description file for AI. It is a community proposal and cheap to add.
For step-by-step checks and what each AI crawler does, see make your site readable by AI.
Step 5: Track and adjust
After publishing, ask the same Step 1 questions again and compare before and after. AI has to index a new page before it can cite it, so keep measuring weekly and compare after about 4 weeks.
The most direct signal is your new article appearing among the sources AI cites. If nothing improves, rule things out in order: not indexed (back to Step 4), title not matching the question, content not citable enough (back to Step 3), or too much competition (back to Step 2 for a narrower topic).
Common GEO misconceptions
| Misconception | Reality |
|---|---|
| GEO replaces SEO | AI often uses search engines first; SEO is the foundation of GEO |
| Asking AI once tells you the result | Answers vary; ask several times and watch trends over weeks |
| More keywords are better | Keyword stuffing scored worse than doing nothing in experiments |
| Only big brands get recommended | Lower-ranked sites gained the most after optimization |
| New pages get cited right away | Changes usually take several weeks to show |
| AI Overviews needs special technical work | Google says there are no additional technical requirements |
Frequently Asked Questions
Q: What is GEO?
GEO is Generative Engine Optimization, the practice of getting AI tools such as ChatGPT, Gemini, Perplexity and Google AI Overviews to mention you and cite your pages.
The term comes from a 2023 paper by Aggarwal et al., published at KDD 2024. It is not a new technology but a set of content and website habits that make your pages easy for AI to find, understand and cite.
Q: How do you do GEO?
GEO is a five-step loop: measure, pick topics, write, make your site readable to AI, and track and adjust.
Step 4, making your site readable, is a site-level setup done once; the other four steps repeat for each topic. The first time, check Step 4 first.
Q: Do I have to choose between GEO and SEO?
No. Do both; SEO is the foundation of GEO.
Google says a page must be indexed and eligible to show a snippet to appear as a supporting link in AI Overviews. AI tools also often search the web before answering, so a page search engines cannot find will not be cited.
Q: Where do AI answers come from?
AI answers come from training data and retrieved search results; GEO mainly influences the retrieved results.
Training data only updates with the next model version and carries no source links. Retrieved results are pages the AI reads when it answers; a page can be used as soon as it is indexed, and it comes with a source link.
Q: What kind of content is more likely to be cited by AI?
Content that answers directly at the top, uses specific numbers and cites verifiable sources is most likely to be cited.
In the GEO paper's experiments, citing sources, adding statistics and adding quotations improved visibility by roughly 30–40% over unoptimized pages. Keyword stuffing scored 17.8, below the unoptimized 19.5.
Q: Can small websites get recommended by AI?
Yes, and experiments show lower-ranked sites gain more.
The GEO paper found that with sources added, a site ranked 5th in search results gained 115.1%, while the site ranked 1st dropped 30.3%. How specific and credible your content is can partly make up for a smaller site.
Q: How do I find out whether AI mentions my brand today?
Ask several AI tools a fixed set of questions that do not include your brand name, on a regular schedule, and record the answers.
Ask each question at least 3 times, count how often you are mentioned, and note the position, type of mention, tone and cited sources. Turn on web search when testing so you can see which pages AI cites.
Q: How long does GEO take to show results?
It usually takes about 4 weeks to see a change.
AI search has to index a new page before it can cite it. Keep measuring weekly and compare before and after at about 4 weeks; a single week can be misleading because AI answers vary randomly.
Start monitoring your brand's AI visibility
Free sign-up, 5-minute setup, an automatic BVS report every week
Start free