AI citation tracking

AI citation tracking measures how five leading AI answer engines respond when someone asks a question your business should be the answer to. For every tracked query, CiteFlow records whether your domain appeared, whether it was cited with a link, where it appeared in the response, and which competitors were named.

This is the measurement layer for large language model optimisation (LLMO). Search engines told you what people typed; AI citation tracking tells you what AI assistants actually said back.

What does AI citation tracking do?

For each query you track, CiteFlow runs a check across all five platforms on a weekly cadence and writes a result row per platform capturing:

• Appeared, your domain or brand name was mentioned anywhere.
• Cited, a clickable link to your site was included.
• Citation URL, the specific page that was linked.
• Position in response, bucketed as early, middle, or late.
• Response snippet, the surrounding text where you appeared.
• Competitor mentions, other domains the AI cited for the same query, with snippets. This is valuable competitive intelligence.
• Model version, recorded so you can correlate movement with upstream model releases.

Which platforms are covered?

CiteFlow queries five sources directly through their APIs. No competitor we are aware of currently covers all five.

What's the difference between cited and mentioned?

These are not the same thing, and the distinction matters.

• Cited = the response contains a hyperlink to your site. A user can click through. This is the goal.
• Mentioned (appeared: true, cited: false) = your name or domain appeared in the text but no link was rendered. The user has to type your name to reach you. Better than nothing, but not traffic.

The dashboard breaks these out separately so you can see whether you are earning links or merely getting name-checked.

How does AI citation tracking work?

When you add or refresh a query, three platforms (Perplexity, Gemini, Google AI Overview) run inline in 10 to 15 seconds and populate immediately. The two slower platforms (OpenAI, Anthropic) take a few minutes; CiteFlow shows them as pending while the cron worker resolves them, so the UI never blocks on a slow upstream.

After the initial load, the cron worker re-checks every query weekly across all five platforms and writes a fresh row per platform.

• Cadence: 7 days per query.
• Worker tick: up to 10 due queries plus up to 30 pending placeholders per tick.
• Retries: 2 attempts with linear backoff per platform.
• Cost cap: tier-based monthly spend cap across all platforms. When reached, further checks pause until the next month.

What does a query look like?

Tracked queries should be question-form, not keyword-form. AI assistants are asked questions; SERPs are queried with keywords.

The simplest test: would a human actually type this into ChatGPT? If not, rewrite it.

Why your personal experience may differ

AI assistants personalise responses based on the conversation context, your prior interactions, your account, your location, and (for some platforms) your subscription tier. The same query at the same time can yield different answers for different users.

CiteFlow queries each platform's API with a clean session and a single prompt to give you a consistent, non-personalised baseline. Your customers' actual experience may differ slightly. Treat CiteFlow's numbers as the directional truth (citing or not, trending up or trending down), not a per-character transcript of any specific user's session.

Why "Not mentioned" everywhere is the correct baseline

New sites should expect zero citations across all five platforms for the first month or two. AI assistants cite sources their model trusts; trust takes time to build through:

• Genuine published content covering the query topic.
• Schema markup that helps assistants understand who you are.
• Backlinks and mentions across the open web (so your domain shows up in retrieval results).
• Time, simply, for crawlers and refresh cycles to see new pages.

Use the baseline of "not yet cited" as the starting line, not a failure. The point of the tool is to watch that change.

What are the tier limits?

How is cost calculated?

Each check fires once per platform per cycle. Default per-call costs:

One full weekly cycle across all five platforms costs roughly £0.13 per query. These costs are included in your subscription; we do not pass them on. The monthly cost cap simply protects you from runaway costs if something goes wrong upstream.

How do you bulk-import queries?

For Enhanced and Marketplace tiers, bulk-add queries through the same flow as keywords. The importer deduplicates against existing queries and fits whatever your tier cap allows.

How do you choose good queries?

• Question-form, not keyword-form (see above).
• Match real intent. Track questions your sales team actually hears, not questions you wish you sold against.
• Cover the funnel. Mix awareness ("what is X?"), comparison ("X vs Y?"), and decision ("how do I choose an X?") queries.
• Watch your competitors. The competitor-mentions field tells you which rivals are getting cited; track the queries where they are winning.

Frequently asked questions

Why does my site never get cited? The most common causes, in order of frequency: insufficient content depth on the queried topic; the page is not indexed yet; the page lacks structured data that helps AI assistants identify what it is about; the site lacks authority signals (backlinks, brand mentions) so it does not surface in the retrieval step that precedes generation.

My competitor is cited everywhere. How long until I catch up? There is no universal answer, but customers who publish consistent, structured content covering tracked queries typically see their first citations within 30 to 60 days, and meaningful movement within 3 to 6 months.

Can I add my own custom AI platform? Not today. The five covered platforms account for the overwhelming majority of generative-answer traffic. Adding a niche platform would dilute the spend cap without proportional intelligence value.

How do you troubleshoot citation issues?

A platform stays on pending for more than an hour. The cron tick runs every 5 minutes; pending rows resolve on the next tick after the upstream returns. If you still see pending after an hour, contact support so we can inspect the worker logs.

A platform shows check_unavailable. The provider returned a clear "cannot answer" or rate-limit signal that is not a transient failure. The next cadence cycle will retry from a clean state.

I added a query and only 3 platforms have results. Expected. The remaining 2 (OpenAI and Anthropic) populate within a few minutes via the cron worker. Refresh the page after 5 to 10 minutes.

Related

• Keyword tracking, track Google positions on the same site.
• Performance dashboard, citations feed the composite AI Visibility Score.