9 GEO mistakes that keep AI engines from seeing your site
AI engines fail to see a site for nine typical reasons: from AI crawlers blocked in robots.txt to company data that's different in every place. Almost all of these mistakes show up in a single evening of checking, and most can be fixed on the existing site, without a rebuild. Below, for each mistake: how to check it and what to do.
What does it mean that "AI engines don't see the site"
When ChatGPT or Gemini answer your customer's question, they draw on two sources: what's left in the model after training, and what they found in search just now. ChatGPT with search relies heavily on the Bing index, Gemini and Google's AI Overviews run on the Google index, and Perplexity shows links to the sources it used to compile the answer.
So "don't see" almost always means one of three things. The page wasn't crawled. It was crawled but not understood. It was understood, but your site lost to others on specifics and verifiability. The nine mistakes below fall into exactly these three groups.
Mistakes 1 and 2: the site is technically blocked or unreadable
Mistake 1: AI crawlers are blocked in robots.txt. This is often a leftover from a developer who blocked bots "just in case," or a CDN-side protection setting. If GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, or Google-Extended are under Disallow, your text physically can't make it into the sources an answer draws on.
How to check: open your-site/robots.txt and look for these names. Separately, check whether the site serves bots a captcha or a 403 error: this happens with aggressive bot protection and with country-based blocking. A line-by-line breakdown of the rules is in the guide on robots.txt and AI crawlers.
Mistake 2: an empty technical foundation. We check a site against eight factors: the site responds, there's a title, there's a meta description, there's Schema.org markup in JSON-LD format, there's FAQ markup, there's an llms.txt, AI crawlers are allowed in robots.txt, and there's a sitemap.xml. Checking each item takes about a minute.
To be honest about llms.txt: it's a format proposed in 2024 (llmstxt.org), it's not an official standard, and support from major AI engines isn't guaranteed. The file is quick to make and doesn't hurt, so we add it. But the result doesn't rest on it.
Mistake 3: the site doesn't answer customers' questions
The most costly mistake. The site talks about the company and lists services, while the customer asks the AI in different words: "how much does a crown cost," "can I return an item without a receipt," "what do I do if I flooded my neighbors." These phrasings don't exist on the site, so there's nothing to cite.
How to check: write down ten questions you were asked in chats and phone calls this week. For each one, look for a page on the site where the answer is given in plain text in the first two or three sentences. If there's no such page, that question is closed off to the AI.
How to fix it: one page per question. The heading repeats the customer's phrasing, the first paragraph gives the answer, and what follows covers terms, numbers, and the steps to take.
Want to know where your website stands today? Free: an AI-readiness check, a 30-day plan and your first article in 3–6 minutes.
Get my planMistakes 4 and 5: the key info is hidden in images, and there are no facts
Mistake 4: the key information isn't in the text. Pricing as an image, delivery terms in a PDF, answers to questions in video and stories, prices loaded by a script after a click. A person will figure it out, but a crawler will download the page and find no words.
How to check: select and copy the text from a key page. If the heading copies but there are no numbers, you're looking at an image. A second method: open the page's source code and search it for a price, address, or specialist's name.
Mistake 5: no verifiable facts. An AI engine avoids recommending something it has nothing specific to say about.
| What the customer asks | What the site needs | Common mistake |
|---|---|---|
| How much does it cost | Starting prices or price ranges by service | "Price on request" |
| Are you near me | Address, neighborhood, service area, hours | Only a contact form |
| Who will do the work | Names of doctors or specialists, experience, photos | "A team of professionals" |
| How fast | Timelines by stage | No timelines at all |
| Can I trust you | Reviews on Google Maps, case studies with numbers | Reviews as screenshots |
Mistakes 6 and 7: data is inconsistent, and no one talks about you
Mistake 6: company data differs from place to place. One phone number on the Google listing, another on the site, an outdated address in a directory, the name spelled differently on social media. To a machine, that's a signal the data can't be trusted, and a competitor with consistent contact info gets the answer instead.
How to check: pull the name, address, phone number, and hours into a table from every place you appear: your site, Google Business Profile, maps, directories, social media, marketplaces. Fix any discrepancy down to a single version, including how the phone number is formatted.
Mistake 7: no one talks about you anywhere except your own site. You can see this clearly in our own case. After launching the GEO setup on insync.com.ua, the AI knows the agency and describes its positioning for four out of four branded queries, but for category queries like "best agency" it doesn't name us yet. Your own site covers branded questions; category questions are covered by external mentions: your Google listing, directories, reviews, publications. The reasons are covered in the guide why ChatGPT recommends competitors.
Mistakes 8 and 9: expecting results in a week, and not measuring
Mistake 8: unrealistic expectations on timing. Indexing new pages in Google usually takes 2–3 weeks, and the first mentions in AI engine answers usually show up after 1–3 months. It can happen faster: on our own site we went from zero to 24 citations in ChatGPT in about a week and a half after launching the setup. That's our case, and no one can guarantee "top spot in ChatGPT."
The picture in search is similar. insync.com.ua collected 183 Google impressions from April through June 2026, and by October 23, 2026 that was already 17,000 impressions and 162 clicks over 90 days. The timeline by step is covered in the guide how long GEO takes.
Mistake 9: nobody measures anything. Without measurement, you can't tell what worked, and "this doesn't work" becomes a guess. Once a month, ask AI engines the same questions your customers ask, record who gets named, and check whether your site shows up among the sources.
We've automated this routine in GEO Autopilot: eight articles a month built around your customers' questions, auto-publishing to a WordPress blog, a monthly measurement across AI engines (currently Gemini and Claude, with more to follow), and a dashboard showing the plan, articles, measurements, and payments. A fixed price after a free site review, no minimum term, cancel with one click.
Checklist: how to check all nine mistakes in one evening
Go through the list top to bottom; each item takes 5–10 minutes.
- robots.txt: AI crawlers aren't blocked, and there's no captcha or 403 for bots.
- All eight technical factors are covered, including Schema.org and sitemap.xml.
- There are pages with direct answers to ten real customer questions.
- Prices, address, and terms are written as text, not as an image or in a PDF.
- The site has starting prices, address, hours, service area, and specialists' names.
- The name, address, phone number, and hours match across all sources.
- There are external mentions: a Google listing, directories, reviews, publications.
- The timeline for checking results is set at 1–3 months.
- Measurement across AI engines is recorded and repeated once a month.
If checking by hand doesn't appeal to you, start with a free AI visibility review: enter your site's address and within 3–6 minutes you get a check against eight technical factors, a list of who AI engines and search sources recommend instead of you for your customers' questions, a plan of eight articles over 30 days, and a first article already written for your site. No payment, right on the site. How to check visibility yourself is covered in the guide how to check whether ChatGPT sees your business.
FAQ
Which mistake should I start with if I'm short on time?
Start with the technical one. Open robots.txt and check whether GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, and PerplexityBot are blocked: a blocked crawler zeroes out any content, because the page simply never gets downloaded. Next, check the title, meta description, Schema.org, and sitemap.xml.
Do I need to rebuild my site to show up in AI answers?
Usually not. Almost all nine mistakes can be fixed on an existing site: robots.txt settings, markup, new pages with answers, consistent contact info. A new site is only needed when the platform won't let you add pages, markup, and files to the root.
What if the site is on Wix, Tilda, or Squarespace?
Auto-publishing doesn't work on these builders, so, to be honest, GEO Autopilot isn't a fit for them. For these platforms we have a "Site + GEO" format: a site built in 3–5 days, followed by ongoing management. On a custom-coded site, Shopify, or Webflow, articles arrive in the dashboard as ready-made HTML.
How long should I wait for results after fixing the mistakes?
Indexing new pages in Google usually takes 2–3 weeks. The first mentions in AI engine answers usually show up after 1–3 months, with branded queries getting covered before category queries. A position in a ChatGPT answer can't be guaranteed, which is why we rely on regular measurement.
Do I need llms.txt if it's not an official standard?
llms.txt was proposed in 2024 and remains a community initiative; support from major AI engines isn't guaranteed. The file takes a few minutes to make and doesn't hurt, so we add it. But betting your results on it would be wrong: what actually decides the outcome is crawler access, answers to customer questions, and external mentions.
How do I know the mistakes have actually been fixed?
Check the technical side against the same eight factors: all of them need to be covered. Check the content side with questions: ask AI engines your branded and category queries and record who they name and which sources they cite. Repeat once a month to track the trend.