Most sites that never get cited by AI search fail for one of three reasons: retrieval crawlers are blocked from reaching them, the answer is buried mid-page instead of stated directly, or the content is too thin to answer anything completely. Off-site authority matters too, but it is rarely the first problem — and unlike the others, it is not something you can fix this week.

Here are the causes ordered by how often they turn out to be the real one, with how to detect each.

1. The retrieval crawlers are blocked

This is the most common cause and the most frustrating, because everything else can be perfect. Two distinct categories of crawler exist, and businesses routinely block the wrong one:

  • Retrieval crawlers — OAI-SearchBot, ChatGPT-User, PerplexityBot — fetch pages to answer live questions. Blocking these removes you from AI answers entirely.
  • Training crawlers — GPTBot, Google-Extended, CCBot — affect model training. Blocking these has no bearing on whether you can be cited today.

Plenty of sites intended to opt out of training and blocked retrieval as collateral damage. How to detect: check robots.txt for the named agents, not just the wildcard. Also check CDN and bot-protection settings, because those blocks never appear in robots.txt.

2. The answer is buried

An extracted passage is shown without the page around it. Content that opens with context-setting and reaches its conclusion in paragraph four cannot be quoted, because no single paragraph contains a usable answer.

How to detect: take any section heading on your page, read only the paragraph directly beneath it, and ask whether it answers the heading. If it does not, that section is not extractable — regardless of how good the whole section is.

3. The page is too thin to answer anything

A 200-word service page that describes what you offer without answering any question a buyer has is not a candidate for citation. There is nothing to extract. This is especially common on commercial pages written entirely as positioning.

How to detect: list the questions a buyer asks before purchasing. Check how many your page answers specifically, with a number, a timeframe or a concrete constraint. If the answer is none, the page is thin regardless of word count.

4. Your business facts disagree with each other

Assistants assemble a picture from every source they can reach. When your site, your listings and your profiles describe your business differently, the uncertain facts get omitted rather than repeated — because repeating a contested claim is riskier than leaving it out.

How to detect: search your business name and compare what each result says about what you do, where you operate and how to reach you. Inconsistencies you have stopped noticing are exactly the problem.

5. The structured data is broken or dishonest

Two separate failures here. Broken markup — a trailing comma, an unescaped quote — causes the entire JSON-LD block to be discarded silently. Dishonest markup, such as FAQ schema describing questions the page does not visibly show, is a policy violation that makes the source look unreliable.

How to detect: validate that every JSON-LD block parses, then confirm each claim in it corresponds to something a visitor can actually see.

6. The content only exists after JavaScript runs

If the meaningful content is injected client-side, some crawlers will see an empty shell. Search engines have improved at rendering; retrieval crawlers vary, and several do not execute JavaScript at all.

How to detect: view the raw page source rather than the inspector, and look for your actual body copy. If it is not there, some systems will never see it.

7. Nothing external corroborates you

The slowest problem to fix and the least likely to be your first one. Sources that other reputable sites reference are safer to repeat. If nothing anywhere corroborates your existence or expertise, you are a riskier citation than a competitor who is referenced widely.

How to detect: this one is genuinely hard to self-assess, which is why it should be the last thing you investigate rather than the first thing you blame.

Fix them in this order

  1. Crawler access — binary, fast, and blocks everything else.
  2. Broken or dishonest structured data — a correctness problem, not an optimisation.
  3. Buried answers — editing existing content, not writing new content.
  4. Thin pages — real work, high return.
  5. Entity consistency — tedious, cheap, quietly important.
  6. Off-site authority — slow, ongoing, and last.

The first three are usually a day of work between them, and they are the difference between being ineligible and being in the running.

Check your own page

Our free AI search readiness checker tests causes one, two, three, five and six automatically on any public URL and tells you which apply. Causes four and seven need human judgement — that is what SEO & AEO services covers. For the prescriptive version of all this, see how to get cited by AI search.

Frequently asked questions

How do I know if AI crawlers can reach my site?

Check robots.txt for rules affecting OAI-SearchBot, ChatGPT-User and PerplexityBot specifically, not just the wildcard rule. Then check that nothing at CDN, firewall or bot-protection level is rejecting them, because those blocks do not appear in robots.txt at all. An automated checker will test the robots.txt half in seconds.

Does having few backlinks stop us being cited?

It contributes, because corroboration across sources is one of the signals that makes a claim safe to repeat. But it is rarely the first problem. Blocked crawlers, thin content and answers buried mid-page are far more common causes, and all three are fixable without any outreach.

Our competitor gets cited and their content is worse. Why?

Usually because theirs is easier to use. They answered the exact question in one self-contained passage while yours spread the answer across three sections, or their business information is consistent everywhere while yours differs between your site and your listings. Being better is not sufficient if being useful is harder.

How long after fixing things will we get cited?

There is no reliable timeframe, and anyone offering one is guessing. Pages have to be recrawled and reprocessed, and citation depends on questions actually being asked where you are the best available source. Treat it as a gradual improvement in eligibility rather than a switch that flips.

Ask Devoq for a manual review if the automated checks come back clean and you still are not appearing.