Your site has traffic. AI engines still ignore you. Here is how to find out why.

Why do sites with Google traffic get ignored by AI engines?

Because Google rankings and AI answer selection are different judgments over overlapping inputs. Ranking rewards a page's standing for a keyword; an AI answer selects sources that directly support the specific sentences it is generating, mixes in independent sources like reviews and threads, and attributes claims to a business entity it has to resolve confidently.

A page can clear the first bar and miss the second for reasons that are invisible in your analytics. The good news: the causes sort into six verifiable classes, and testing them in order finds yours. The full model behind the stages is in the getting cited by AI pillar.

1. Is the page indexed and accessible?

Check access before anything else, because a blocked page makes every other diagnosis meaningless. Google traffic proves Googlebot access, and nothing else: ChatGPT search depends on OAI-SearchBot, and Perplexity on PerplexityBot; either can be blocked by a robots.txt template or a CDN bot rule while Google sails through.

Verify the three layers: robots.txt rules per search crawler, what the CDN/WAF actually serves them, index presence per platform. The crawler-by-crawler method is in which AI crawlers should your website allow. If this class fails, fix it and re-baseline before judging anything downstream.

2. Is the important answer actually on the page, clearly?

Read the candidate page as an outsider and ask: is the direct answer to the buyer's question stated in text, near the top, in plain sentences a machine could quote? Pages that rank often accumulate content that talks around the question: category context, feature narratives, testimonials, with the actual answer implied rather than stated.

Also check the rendering: content that only exists after JavaScript runs, or inside images, may be absent from what non-browser readers receive. The blueprint for fixing this class is how to structure a page so AI can cite it accurately.

3. Does the page match the question's intent?

An engine matching "best accounting software for construction" wants a page whose type matches the intent: a comparison, a category page, an answer to that question. If you only publish feature pages, you can rank on brand strength while offering nothing an answer engine can select for comparison or recommendation intent.

Map your lost questions to page types: informational questions need direct-answer content, comparison questions need honest comparison pages, local service questions need location-and-service pages. A missing page type is a content-strategy gap, not a quality gap, and no amount of polishing the wrong page fixes it.

4. Is the business entity ambiguous or inconsistent?

Ask each engine "what is [your business]?" and read literally. If the answer is wrong, vague or confuses you with someone else, the engine cannot confidently attribute your pages to a nameable business, and mentions and recommendations fail even where citations might succeed.

The usual causes are mundane: name variants across site, profiles and directories; a thin About page; missing or conflicting Organization/LocalBusiness schema; unclaimed profiles carrying stale facts. The repair sequence is in how AI engines identify your business.

5. Do independent sources support competitors instead?

For each lost question, list the URLs the engines cited and check each: does it mention competitors, and does it mention you? When answers are grounded in review platforms, directories, comparison articles and threads that describe competitors only, the engine recommends competitors regardless of your on-page quality.

This class is the most common explanation for the specific pattern "our pages get cited sometimes, but the recommendation always goes elsewhere". The evidence-first playbook for closing named source gaps is in why reviews, directories and communities influence AI recommendations.

6. Is the test itself unstable?

Before concluding you are invisible, check the measurement. One-off checks of one engine with ad-hoc phrasing produce unreliable verdicts: answers vary run to run, engine to engine and region to region, and a differently phrased question is a different question.

A sound test uses a stable set of real buyer questions, the same engines and regions every time, and repeated runs to establish variance before judging. If your "no citations" conclusion came from a handful of manual checks, run a real baseline first; the design is in how to measure whether an AI visibility fix worked.

A practical sequence

Work the classes in order, one change at a time, re-testing between:

1. Access first, because it gates everything and is fast to verify.
2. Answer presence and intent match next, on the handful of pages mapping to your most valuable lost questions.
3. Entity in parallel with content, since records correction is independent work.
4. Independent proof once your pages are eligible, targeting the specific sources cited for your questions.
5. Measurement discipline throughout: baseline, ship, re-test, attribute.

Resist fixing all classes at once on all pages: an unattributable improvement teaches you nothing for the next hundred questions.

Where CiteAgentic fits in this loop

Product example. A CiteAgentic scan runs this diagnosis per tracked question: technical checks cover crawler access and indexability, the page audit checks whether the mapped page answers the question directly, entity checks review your records for consistency, and the citation evidence shows which sources each engine cited and where competitors appear without you. Each finding becomes a recommendation tied to the affected questions, drafts are prepared for human review with unknown facts marked, and the same questions are re-tested on later scans with page state, validation state and AI outcome reported separately. No finding claims a guaranteed citation; the re-test reports what actually changed.