A lot has been written about GEO, but most of it stops at "What is GEO?" This article focuses on practice: what work does a brand need to do, and in what order, to be mentioned more often in AI answers?
What is GEO, in short?
GEO stands for "generative engine optimization." It means optimizing for generative AI engines. The goal is for your brand or your content to be mentioned or cited in the answers of assistants like ChatGPT, Gemini and Perplexity.
The term was introduced in a paper published in 2023 by researchers from Princeton and other institutions and accepted at KDD 2024. The paper shows that, in a test environment the authors built themselves, content edits could increase visibility in AI answers by up to 40%. The same paper also stresses that which method works depends on the domain. In other words, there is no single recipe that fits everyone.
How it relates to SEO: GEO does not replace SEO. Google says that AI Overviews and AI Mode require no additional requirements or special markup; basic SEO work applies here too. GEO adds one more question on top: "Which pages does AI read, and does it mention me?"
Step 1: Write down your buyers' questions
GEO work starts with a list of questions, not a list of keywords. What do your buyers type into an assistant?
- Category questions: "What are the best X companies?"
- Local questions: "Recommend a company that offers X service in Bursa."
- Needs-based questions: "Which X is right for a small team?"
- Comparison questions: "Is X or Y the better fit?"
Your brand name should not appear in the questions. A question that includes your name measures awareness, not visibility. Starting with 10 to 25 questions is enough.
Step 2: Measure where you stand today
Ask the questions to ChatGPT, Gemini and Perplexity separately and record whether your brand is mentioned. Answers change every time. SparkToro's research shows that the chance of two answers to the same question giving the same list of brands is under 1%. So do not look at a single answer; look at the mention rate across many questions and repeated runs.
This first measurement is your starting point. It is how you tell whether each later step worked.
Step 3: Check that bots can access your site
If AI engines cannot get information about your site, your pages cannot be cited, no matter how well you write them. Check the following in your robots.txt file:
- For ChatGPT, OAI-SearchBot. According to OpenAI's documentation, sites that block this bot are not shown in ChatGPT search answers. There is a separate bot for model training (GPTBot); you can configure the two separately.
- For Perplexity, PerplexityBot. Perplexity's documentation recommends allowing this bot if you want to appear in search results.
- For Google, the regular Googlebot. The Google-Extended setting, according to Google, relates to training Gemini models and use in Gemini products; it does not affect inclusion in Google Search.
Besides robots.txt, you should also review your sitemap, page titles, meta descriptions and canonical tags. These are classic SEO checks, but they matter just as much for AI.
Step 4: Find the external pages the engines read
AI answers are often built from other sites. In an Ahrefs study on ChatGPT, 43.8% of the cited links were "best X" style blog lists.
What to do in your own category:
- Collect the pages cited in the answers.
- Sort out which ones are list pages and which are guides, forums or marketplaces.
- Flag the lists your competitors are on and you are not.
- For editorial lists, reach out to the page owner or writer. In a short, specific email, explain your brand, what you do and why you fit the list.
Your chances of getting onto a "best companies" list on a competitor's own site are low. It is more efficient to put your effort into independent editorial lists.
Step 5: Shape your own pages around buyer questions
A few simple rules for page text:
- In the first paragraph, clearly state what you do, who you do it for and where you operate.
- Write subheadings in the form of the buyer's question and put a short answer right below them.
- Describe your services and products in text, not inside images.
- Do not add numbers you do not have or claims you cannot verify. This both damages trust and spreads wrong information.
Before spending time on a special file or markup, remember what Google has said on this: AI Overviews and AI Mode do not need a special schema or an AI text file. The real difference comes from the page giving a clear answer to the question.
Step 6: Measure again with the same questions
GEO is not a one-time job. After making changes, measure again with the same question list. A brand's overall standing changes slowly; weekly tracking is enough for most brands. Do not change the questions every week, or you will not be able to compare the results.
Keep a simple change log: date, the work done, the related questions. Measure the same question list for three months and read it next to that log. If the rate rises after a piece of work, that is a good sign, but on its own it does not prove the work caused it; answers also change on their own. How the numbers are calculated is on the methodology page.
Metrics to track: the share of answers in which your brand is mentioned, your share of all brand mentions in the answers, and how your competitors do over time.
What are the most common mistakes in GEO work?
There are a few points to watch that come up often in practice:
- Starting without measuring. Without a first measurement, you cannot know whether a change worked.
- Deciding based on a single answer. Changing your whole site because you did not appear once creates unnecessary work. Appearing once and relaxing is also misleading.
- Focusing only on your own site. Engines also read external pages. If you are not on list and guide pages, fixes on your own site will not close the gap by themselves.
- Putting too much effort into cosmetic work. Special files and complex markup do not replace a page that clearly answers the question.
- Adding made-up information to your text. Phrases like "industry leader" or unverified customer counts look tempting in the short term, but wrong information can carry over into AI answers too.
- Changing the questions often. If you measure with different questions every week, you cannot see a trend.
Can you do GEO yourself?
Yes. All of the steps above can be done with a spreadsheet, some time and regular tracking. As the number of questions and engines grows, the measurement and source collection take time; this is where a tool makes the job easier.
AIShortlist brings this working order into one dashboard: it suggests questions from your site and does not measure until you approve them, measures the three engines separately, shows the cited list pages and where you are missing, runs a technical site check and suggests page text using only information found on your site. With the free trial, you can run your first measurement without giving credit card details.
One last reminder: GEO is work that raises your chances of being recommended, not a guarantee. Be careful with anyone who promises that.