How long after fixing something should I expect to see a change?

Days to weeks for retrieval-driven engines, because they fetch at query time. Longer for anything leaning on training data. Technical fixes surface fastest; authority and corroboration changes are the slowest.

Could it just be that my competitor is bigger?

Domain authority matters, but it is rarely the whole story and it is the least actionable explanation, which is why it should be your last hypothesis rather than your first. Smaller sites out-cite larger ones regularly by being more specific and better structured on a narrow question.

What if I am cited by some engines but not others?

Normal and diagnostic. Engines use different retrieval backends and citation thresholds. Per-engine divergence usually points to retrieval configuration or freshness rather than to content quality, since your content is identical across engines.

Is there a single fix that works most often?

Yes, in a boring way: making sure AI crawlers actually receive your content. It is the most common cause and the cheapest to fix, and it is skipped remarkably often because it is not interesting.

Why Your Brand Is Not Being Cited

Diagnose in order of probability

The instinct on discovering you are absent from AI answers is to rewrite content. Content is usually not the problem, and rewriting is expensive.

Below are nine causes in rough order of how often we see them discussed in published audits and vendor case studies, which is also close to their order of cheapness to fix. Work down the list. Stop when you find yours.

1. AI crawlers are blocked

The most common cause and the least interesting, which is exactly why it survives.

Check your live robots.txt, not the version in your repository or your memory. Look for a wildcard Disallow: /, an explicit AI-crawler block added during the 2023–24 publisher blocking wave, or a directory-level Disallow covering your pricing or documentation.

Test: read the file. Fix: allow GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. Time to see change: days.

2. Content only exists after JavaScript runs

Second most common, and structural.

Some AI crawlers do not execute JavaScript; those that do give it a smaller budget than Googlebot. A single-page application that renders content on hydration is frequently an empty shell to them.

Test:

curl -A "GPTBot" -sL https://example.com/pricing | grep -c "99"

If your prices, features and specifics are not in that HTML, they do not exist as far as those engines are concerned.

Fix: server-render or statically generate any page you want cited. Time: an engineering cycle, then days.

3. Bot mitigation is blocking the crawl

Configuration says yes, the edge says no. WAFs and bot-management products routinely classify AI crawlers as hostile automation and serve challenges, 403s or rate limits.

Test: filter CDN logs by AI user agent and read the status-code distribution. Fix: add AI crawlers to your verified-bot allowlist and check rate limits against burst behaviour. Time: minutes to configure, days to see effect.

4. The answer is buried

Now content becomes relevant — but structurally, not qualitatively.

Retrieval scores passages. A page that reaches its answer in paragraph eleven, after context-setting and narrative, offers no strong chunk. A competitor stating the same answer in their first sentence wins with worse content.

Test: search a conventional engine for the queries a model would likely generate from your buyer's question. If you rank well there but are never cited, structure is the probable cause.

Fix: answer-first openings, question-phrased headings, self-contained sections. See Structuring Content So AI Can Quote It. Time: hours per page, then weeks.

5. Duplicate coverage splits your own authority

A cause teams create for themselves.

Four pages each partially covering a topic hand the re-ranker four mediocre candidates. The competitor with one comprehensive page gives it one strong candidate and wins. Your pages also compete with each other, so none accumulates the internal links or authority it would have had alone.

Test: search your own site for your key topic. If several pages plausibly answer the same question, that is the problem.

Fix: consolidate into one canonical page and redirect the rest. Painful, effective, and it usually improves conventional rankings too. Time: weeks.

6. Nobody else corroborates you

Models weigh consensus. A claim only you make gets hedged, dropped, or attributed to the general view rather than to you.

Test: ask an engine your category question and read which domains are cited. If review sites, directories and community threads dominate, your problem is off-site.

Fix: get accurate, current information into the sources being cited. Complete your directory listings. Earn independent coverage. Participate where your buyers actually discuss the category. This is slow and it compounds. Time: months.

7. Your facts are stale or wrong in the model's view

Common, damaging, and the fastest thing to fix once found.

Models frequently describe vendors using outdated information: a superseded price, a feature shipped last year reported as missing, a limitation resolved two releases ago. This is not a citation problem — it is worse, because you are being actively misdescribed to buyers.

Test: ask several engines directly what your product costs, what it does, and what its limitations are. Read the answers carefully against reality.

Fix: publish an unambiguous, well-structured page stating the correct facts, add Product/Offer and FAQPage schema, and push the correction into the third-party sources being cited. Then monitor, because corrections need to propagate before they take. Time: days to weeks, and it is the highest return-per-hour fix in AEO.

Finding these systematically rather than by chance is the argument for false-claim monitoring; CiteCue's Brand Risk surfaces incorrect statements explicitly rather than leaving you to stumble across them, which matters because you cannot correct what you have not noticed.

8. An aggregator owns your ground

Sometimes the cited source is a listicle or directory page that mentions you in passing. The model is answering from a page about your category rather than from any vendor.

Test: check whether the cited domains are aggregators rather than vendors.

Fix: two fronts. Make your listing on those aggregators complete and accurate, since that is where the model is reading. And build a page that beats the aggregator on specificity for the queries you care about — aggregators are broad and shallow, which is a beatable profile if you are narrow and deep.

9. Your pages are hostile to agents

The newest cause, and invisible in conventional analytics.

An agent sent to check your pricing or start a trial fails on things no human tester notices: a price behind a hover state, a plan comparison requiring a dropdown, a signup step depending on a third-party script. When the agent stalls, the user hears that your product is unclear or unavailable. There is no conversion and no complaint, so nothing appears in your funnel.

Test: have an agent attempt your three most important buyer tasks and record where it stops. Doing this properly is what Agent Usability testing automates; almost nothing else in the category tests it at all, which makes it the most under-diagnosed cause on this list.

Fix: put critical facts in plain server-rendered HTML, remove interaction dependencies from key paths, keep signup flows simple. Time: an engineering cycle.

The one to consider last

Your domain is smaller than your competitors'. True often, and the least useful conclusion available, because it is not actionable in a quarter.

It is also frequently an excuse. Smaller sites out-cite larger ones regularly by being more specific, better structured and more current on a narrow question. Authority is a tiebreaker, not a gate. Exhaust the eight actionable causes before accepting this one.

Working the list

Do it in order. Steps 1 to 3 take under an hour and resolve a large share of cases. Steps 4 and 5 are content work with a clear method. Steps 6 to 8 are slower programmes. Step 9 needs tooling.

The mistake to avoid is starting at step 4 because content is the interesting part. Rewriting pages that a crawler cannot fetch is the most common way to spend a quarter on AEO and have nothing to show.