Start with a free AI Visibility Snapshot →
Method

How CoreTxt audits AI visibility: method, scoring and limits

CoreTxt's Deep Audit asks 9 to 20 customer questions in ChatGPT, Gemini, Perplexity and Claude, records whether your business is mentioned, recommended and cited, notes which competitors and sources appear, and scores nine categories. Every finding is ranked P0 to P2 and labeled by how it was established, so you can tell what was observed from what was interpreted. The score is a directional indicator, not a ranking, and AI answers vary, so a report describes a sample, not a guarantee. CoreTxt measures and recommends. It does not change your website and does not promise mentions, rankings or citations. The free Snapshot uses a separate, published scoring method for technical readiness.

See Deep Audit pricing Start with the free Snapshot

Deep Audit $347 one time. The free Snapshot needs no credit card.

By CoreTxt · Updated November 13, 2026

What does the Deep Audit test?

It tests the questions your customers ask, not only your brand name. A typical set mixes three kinds of question:

  • Unbranded questions, such as "who is a good certified roof coating contractor near me", where the assistant picks who to name.
  • Comparison questions, such as "which flat roof coating is best", where businesses and products are weighed against each other.
  • Brand and detail questions, such as "tell me about" your business or how a process works, where accuracy matters.

You provide the questions or CoreTxt drafts them with you. The Deep Audit covers 9 to 20 of them.

How are the questions run?

Each question is asked in each of the four assistants, and every answer is recorded as one observation. Nine questions across four assistants gives 36 observations. For each one the report records:

  • whether your business was mentioned
  • whether it was recommended, meaning named as an option to choose
  • whether the answer cited a source and which one
  • which competitors were named
  • notes on how the assistant described you, and whether the facts were correct

Answers can change with wording, location, whether web search is on and the model version. A single observation is therefore a snapshot of that assistant on that day. Monitoring plans re-test over time so you can see the trend instead of one answer.

How are findings labeled and ranked?

Every finding carries an evidence label and a priority.

Label Meaning
VERIFIED Directly supported by research or source evidence
AI-OBSERVED Directly seen in an assistant's answer
SEARCH-OBSERVED Directly seen in search research
INFERRED The audit's interpretation of the observed evidence
UNVERIFIED Not independently confirmed

Priorities run from P0 (address first) to P2 (lower impact). Inferred findings state their confidence as high, medium or low, because a cause drawn from a sample is a hypothesis, not a proof.

How is the Deep Audit score built?

The overall score combines nine weighted categories and is labeled directional.

Category Weight
AI Visibility 20%
Entity Clarity 15%
Local Authority 15%
Content / Topical Authority 15%
Third-Party Authority 10%
Technical GEO 10%
Competitive Positioning 5%
AI Accuracy 5%
Measurement Readiness 5%

If a category has no evidence, the report marks it as unscored and says so. It is not counted as zero, and the report notes how much of the total weight was scored. The score should be read as a directional indicator for prioritizing work, not as a complete measure of AI visibility.

See what a Deep Audit report looks like

$347 one time. 9 to 20 customer questions across four AI assistants, a competitor gap, source analysis and a 90-day roadmap, as a PDF within 1 day.

See Deep Audit pricing View the sample report

How is the free Snapshot scored?

The free Snapshot scores technical readiness only, out of 100. It covers technical access (25%), entity signals (37.5%) and content (37.5%), and it does not ask any assistant a question. The full method is on the free GEO audit tool page, and a real result is on the free report page.

What does the audit not cover?

It has limits, and the report states them.

  • It does not test Google AI Overviews. The report names them as a channel to baseline separately if they matter to you.
  • It samples a set of questions at one point in time. It does not show every question a customer could ask.
  • Each report lists what was and was not supplied, such as test dates, engine settings or technical scan output, and does not draw conclusions from missing evidence.
  • It does not implement fixes, and it cannot promise that an assistant will mention, recommend or cite you.

Where can you see what a report looks like?

See the sample report for the format, the case study of our own audit for a real example of the free audit, and the cost page for how the Deep Audit compares with other quotes you may receive.

Frequently asked questions

How many questions does the Deep Audit test?

It tests 9 to 20 customer questions. Each is asked in ChatGPT, Gemini, Perplexity and Claude.

Is each question asked more than once?

Each question is asked in each of the four assistants and recorded as one observation per assistant. Because answers vary, the report treats the results as a snapshot, and monitoring re-tests over time.

Does the score mean I rank in AI?

No. The score is a directional indicator of strengths and gaps across nine categories. It is not a ranking and it does not predict whether an assistant will name you.

What does INFERRED mean in a report?

It means a likely explanation drawn from the evidence rather than something checked directly. Each inferred finding states a confidence level.

Does the Deep Audit cover Google AI Overviews?

No. It covers ChatGPT, Gemini, Perplexity and Claude. The report can recommend a separate AI Overview baseline.

Do you guarantee that AI will recommend my business?

No. CoreTxt measures visibility and recommends actions. No audit can guarantee a mention, a recommendation, a ranking or a citation.

Related: Wrong information in AI answers · Why a business stopped appearing · Monthly monitoring.

See the Deep Audit or start free

See the Deep Audit price and what is included, or check your website first with the free Snapshot. CoreTxt measures visibility and recommends actions. It does not implement fixes, and no audit guarantees mentions, rankings or citations.

See Deep Audit pricing Start with the free Snapshot