AI citation tracking for brands that need evidence

A citation is a visible source reference, URL, or source card used by an AI answer surface. A brand mention without a visible source is recorded separately.

No. An answer can cite a page while describing or recommending another brand. That is why TomHow records what the answer actually says, which brand it names, and which URL it cites instead of treating any URL appearance as a recommendation.

The audit can check relevant surfaces such as ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and other answer surfaces where access and evidence allow.

It can show absence from a defined prompt set on a defined date. It cannot prove absence from every possible prompt, location, user state, or platform mode.

Agree the cadence before work starts. Thomas checks the same buyer questions weekly during an active improvement run where platform access allows; monthly reporting then explains the dated observations and their limits. A repeated citation is useful evidence, not a guaranteed or universal outcome.

AI citation tracking measures whether agreed buyer questions produce a brand mention, a linked source, both, or neither on a named answer platform at a recorded time. TomHow keeps the exact question, platform, date, visible answer and cited URL together, because a citation without that context is easy to misread. The record separates a business being named from one of its pages being used as a source. It also notes when an answer is unavailable or cannot be verified. A fixed panel can be repeated to show whether the observed answers changed, but it does not represent every user, prompt variation or future response. It cannot prove that one page edit caused a mention, citation, enquiry or sale. The evidence is used to find weak pages, missing source material and measurement gaps. Available platforms, access and repeat frequency are confirmed in the scope before work starts.

Tracking does not prove trust, rankings, revenue, or future citations. It gives a dated record of what selected AI systems returned for selected prompts. Method source: Thomas Howcroft, founder interview, 11 October 2026; edited for clarity.

OpenAI web crawler documentation →

Search visibility audit GBP 2,250 Standalone diagnostic. No retainer required. 01 A dated search baseline 02 Evidence for each finding 03 A prioritised repair plan Founder-led. Written findings. Clear next steps.

01 / What gets tracked

Each check needs enough context to be useful later. A row that only says visible or not visible is too weak. The useful version records the prompt, platform, date, answer, cited URLs, competitors, and limits.

• Buyer prompts and comparison prompts
• Brand mentions and competitor mentions
• Visible citations and cited URLs where available
• Wrong or incomplete descriptions
• Blocked, inconclusive, or unavailable rows

02 / What it informs

Citation tracking should not sit in a dashboard by itself. The useful question is what the evidence says to change next across SEO, GEO, AEO, content, schema, and internal links.

• Pages that need clearer answers
• Missing proof or corroboration
• Competitor pages that appear repeatedly
• Entity confusion around brand, service, or category
• Prompts that need deeper source material

03 / Founder method

Thomas first asks whether a reader can quickly tell what the page is about and what service it offers. He checks buyer intent and search data, gives the main question a direct answer, then adds useful depth. Repeated prompt checks help him decide whether to improve that page, create a page for a distinct topic, or seek stronger independent proof.

• One main search focus with related buyer questions covered in depth
• Clear headings, relevant internal links and sound page performance
• Proof appropriate to the service, such as evidenced work or independent reviews
• Dated repeat checks that keep mentions and linked citations separate

04 / Proof labels

AI answers vary by tool, location, wording, time, and retrieval mode. TomHow labels evidence so a buyer can see what is verified, directional, anonymised, fabricated sample data, or still a hypothesis.

• Verified when the evidence is visible and dated
• Directional when the pattern is useful but incomplete
• Anonymised when client data cannot be shown
• Fabricated when the page uses a public sample
• Hypothesis when the cause still needs testing

05 / Limits

A brand missing from one answer does not prove the brand is absent everywhere. Blocked rows, no-citation answers, and changing platform behaviour must be labelled properly.

• Blocked rows are not proof of absence
• A mention is not the same as a citation
• A citation is not the same as a sale
• One answer is not a market-wide conclusion
• Scores need interpretation before action

Founder of TomHow, based in Stevenage. SEO and AI search consultancy.

• How it worksThe method that turns evidence into next actions.
• Tool vs managed serviceWhen dashboards need interpretation.
• Citation gap report limitsDistinguish missing citations from unavailable data.