It is one of the more disorienting things to discover. You hold position one for your most valuable term. You have the backlinks, the crawl budget, the green scores. Then you ask ChatGPT the same question your customers ask, and it recommends three competitors and never mentions you.
This is not a bug in the model and it is usually not bad luck. Ranking and being cited are two different competitions judged on overlapping but distinct criteria. Here are the seven reasons the gap opens, roughly in the order we see them.
1. Your answer is buried below the fold of the text
Search rewards a comprehensive page. A generative engine pulls a passage. If your page opens with 300 words of context setting before it says the thing the question asked, a retriever scoring chunks will find a competitor's blunt second sentence more relevant than your carefully built argument.
The fix is structural, not editorial: state the answer in the first paragraph under the heading that asks the question, then elaborate. You lose nothing with human readers, who also prefer this.
2. The page only exists after JavaScript runs
Googlebot renders JavaScript. Most AI crawlers largely do not, or do it inconsistently. A single-page app that ships an empty div and fills it on the client can rank perfectly well and still look like a blank page to GPTBot, ClaudeBot or PerplexityBot.
Check by fetching your own page the way they do:
curl -s https://example.com/your-page | wc -wIf that word count is near zero but the page is full in a browser, you have found your problem. Server-render or prerender the pages that matter.
3. You blocked the crawlers, possibly by accident
Plenty of sites added GPTBot and friends to robots.txt in 2023 and 2024 during the training-data backlash, then forgot. Blocking training is a legitimate choice. Blocking the retrieval crawlers that build live answers is a decision to be invisible in them, and it is often made without anyone realising the two are different.
Audit your robots.txt for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended, and decide each one deliberately.
4. Nothing corroborates you
Answer engines weight agreement across sources. A claim that appears only on your own domain is a claim from an interested party. The same claim echoed on a forum, a review site, a comparison article and a reference page reads as established fact.
This is why off-site work matters more for GEO than it did for classic link building, and why it cannot be faked at volume. See Off-Site GEO for the practical version.
5. The page has no date, author, or entity anchoring
Retrieval systems are conservative about staleness and provenance. A page with no visible publication date, no named author, and no structured data giving the publisher an identity is harder to trust and easier to skip. This is cheap to fix and rarely done properly. See Structured Data for AI Answers.
6. You are optimised for a keyword, not a question
"Enterprise CRM solutions" is a keyword. "Which CRM should a 12-person agency use if we already pay for HubSpot marketing?" is what somebody actually types into an assistant. Prompts are longer, more qualified, and full of constraints. Pages written to rank for a two-word head term frequently answer none of the qualifiers, so they never surface for the conversational form of the same need.
7. Your content is unquotable by construction
Marketing prose optimised to avoid commitment is prose a model cannot use. "Powerful, flexible solutions tailored to your needs" contains no extractable fact. A model composing an answer needs something it can attribute: a number, a limit, a supported version, a price, a named tradeoff. Vagueness is not neutral in GEO. It is disqualifying.
How to confirm which one is biting you
Work through it in this order, because the cheap checks eliminate most cases:
- Fetch the page without JavaScript. Is the content there?
- Check robots.txt for the AI crawlers by name.
- Read your own first paragraph. Does it answer the question, or introduce it?
- Search for your brand plus the topic outside your own domain. Does anyone else say it?
- Ask the engines the buyer questions directly and read the full answers, not just whether your name appears.
That last step is the one teams skip, and it is the only one that tells you what is actually being said. If you want it done continuously rather than as a one-off audit, that is what this platform measures.
The reframe
Position one means you won the competition to be listed. Citation means you won the competition to be quoted. The second one is judged on how easy you are to lift, verify and attribute. Those are writing and markup problems more than they are authority problems, which is good news: they are fixable this quarter.
