AI visibility report
What a Useful AI Visibility Report Should Actually Tell You
An AI visibility report should not be a deck of charts that ends with “improve your authority”. It should make one specific decision auditable: where AI recommends competitors instead of you, what evidence surrounds those wins, what you can reasonably influence, and what should be measured again.
The ten sections, in the order they should appear
This order is deliberately not the order of a typical analytics dashboard. It puts the decision first and the supporting analytics after it.
1 · The decision
The single highest-value finding, in a sentence a non-specialist can act on.
2 · How it was measured
Question set, engines, run count, dates, and the exact provider surface.
3 · Buyer questions
Every question, every engine, recommendation state, and the competitor that won.
4–8 · The evidence chain
Winning-answer themes, cited sources, source addressability, owned-page evidence, product-truth state.
9 · Actions
Each with a target, a change, the evidence behind it, and a confidence bound.
10 · Proof
What was remeasured, what moved, and where the comparison was refused.
| # | Section | The question it closes |
|---|---|---|
| 1 | Executive finding | What is the single most important thing we learned? |
| 2 | Scope and methodology | Could someone else interpret this run later? |
| 3 | Buyer-question table | What happened on every important question? |
| 4 | Strategic positioning | Do the engines associate us with what we want to own? |
| 5 | Competitor intelligence | Who is winning, on which questions, on which engines? |
| 6 | Source and citation evidence | What is the answer leaning on, and can we reach it? |
| 7 | Owned-site evidence | What do our own pages actually say about this? |
| 8 | ProductTruth | Is the claim we are about to make even true? |
| 9 | Best next actions | What should be done first, and on what evidence? |
| 10 | Before/after proof | Did the measured state change after the work? |
Sections 1 and 2 — lead with a decision, then make it reproducible
The report should open with something a person can act on, and immediately show the conditions that produced it.
From the opening finding you should be able to drill into the exact question, the engines, the recommendation state, the competitor, the confidence, the actionability and the next action. A single overall score can sit below that, where it belongs.
Methodology then states the brand and domain, the measurement date, the approved question cohort and its version, the engines and providers, location and language where relevant, the number of runs per question, the web-grounding conditions, the model and provider context, the observation count, and the limitations. If a client challenges a number six weeks later, this section is what makes the answer possible.
Section 3 — the buyer-question table
One rule matters more than the column list: unmeasured providers never enter the denominator.
- the exact approved question, and its intent class
- brand recommendation share, and the strongest competitor
- which engines won and lost
- confidence
- the commercial gap, and the action priority — scored separately
Commercial importance≠Fixability
A gap can be commercially critical and still have no action priority. Merging the two columns is how a report ends up either burying the hard problems or pretending everything is addressable.
Sections 4–8 — the evidence chain
Positioning, competitors, sources, owned evidence and product truth, each with its own discipline.
Positioning
Approved strategic claims classified per engine as won, contested or lost. Usually more useful than share of voice.
Competitors
Recommendation share, which questions each wins, engine differences, newly emerging names — with identity normalised so a cited publisher never becomes a competitor.
Sources
The domains appearing in observed answers, split into relevance and addressability. Never “get listed on all of these”.
Owned evidence
The matched page, its title, the evidence found, the winning-answer themes, and gaps limited to the themes actually observed.
ProductTruth
Verified true, verified false or unknown — checked before any claim is recommended, so optimisation never turns into inaccurate marketing.
Section 7 has one specific wording rule
“No matching evidence was found in the inspected page set” is defensible. “Your website does not have this” is not, because a bounded inspection is not a claim about every page you own.
Sections 9 and 10 — actions, and proof without the overclaim
Every action answers four things; every before/after refuses to become a causal claim.
- Asset or target — what exactly are we working on?
- Change or investigation — what should be done?
- Evidence — what observed answer or page evidence supports it?
- Measurement — which buyer questions will be remeasured?
Watch for this
Avoid “improve content”, “build authority”, “get backlinks” and “add schema” unless each is tied to a specific measured gap. Those four phrases are what a report says when it has not actually found anything.
Before / after≠Causality
The report can state “measured recommendation share changed from 0% to 33%”. If the question, providers or context changed enough to make the comparison invalid, it should refuse the number and explain why. Trust is worth more than always having a delta.
A real example: the founder-owned OwnQR run
We used OwnQR to find out whether the intelligence layer surfaces a genuine commercial finding from live AI answers, or just another chart. These are the actual figures.
| What was measured | Result |
|---|---|
| Buyer questions in the approved cohort | 14 |
| Engines | 3 — OpenAI, Gemini, Perplexity |
| Stored observations | 84 |
| Owned pages inspected | 40 |
| High-intent questions | 14 of 14 |
| Measured losses | 11 |
| Directional gaps | 3 |
| Findings held back as insufficient evidence | 0 |
| Provider cost of the last full re-analysis | $0 — re-scored from stored answers |
A competitor is currently recommended instead of you on 11 of 14 high-intent buyer questions.
Underneath that headline, the positioning section carried the finding that actually mattered: three stated claims contested across engines — recommended by OpenAI, lost to Uniqode on Gemini and Perplexity — and the “editable destination after printing” claim lost on every measured engine.
That is the report shape we are arguing for: which positioning claim, which engines, which competitor, what evidence, what next. Note what is not in that table — no invented revenue impact, no “you are losing $X per month”, and no causal explanation of why Gemini chose someone else. None of those are measurable from observational data, so none of them are in the deliverable.
What not to put at the top of a report
Lead with the highest-value decision and let the analytics support it.
Not this
- twelve charts
- a 0–100 score
- “500 prompts measured”
- the full citation list
- a generic executive summary
What the report is today
A self-contained HTML file, reviewed by a person before it is sent, with the ten sections in that order.
There is no client login, no portal and no scheduled delivery — during a founding pilot the deliverable is a file you can open, keep and forward, plus a call to walk through what it does and does not support. For agencies that is deliberate: a file survives being forwarded to a client's CFO, and a login does not.
Common questions
What should an AI visibility report include?
Buyer-question results, recommendation status per engine, competitors, citations and sources, owned-site evidence, product truth, prioritised actions, methodology and limitations, and historical change once there is a compatible earlier run to compare against.
Is an AI visibility score enough?
No. A score is a summary. The report has to show the questions and the answer evidence behind it, or nobody can audit the number — including the person who has to defend it to a client.
Can a report prove why a model ranked a competitor?
No. It can show observed answer themes, citations, competitor patterns and owned evidence. Those support a diagnosis; they do not expose a hidden ranking algorithm, and a report claiming otherwise is guessing.
How often should a report be generated?
For ongoing monitoring, weekly or monthly depending on the value of the tracked questions and the cost of measuring them. An important completed action can be remeasured on its own without re-running the whole cohort.
Can I use an AI visibility report with clients?
Yes. Preserve the methodology, the evidence and the limitations so the report stays defensible when a client pushes back on a number — which is exactly when a thin report falls apart.
Methodology
Every figure in the worked example above is read from a frozen measurement run, not estimated. The cohort was approved before any provider call was made; each question was measured across all three engines; failed calls were excluded from denominators rather than counted as absence; and the full re-analysis that produced these numbers cost nothing at the providers because the raw answers were retained and re-scored offline.
The brand measured is founder-owned. It is included because it is the only brand whose results we can publish without a customer's permission, and it is labelled that way everywhere it appears.
How we measure, and how this page is sourced
What we actually measure
Every result comes from the official OpenAI, Google Gemini
and Perplexity APIs with web search or grounding enabled. In our data those
surfaces are recorded as openai_api_web_search,
gemini_api_google_search and perplexity_api_sonar, and reports name
them explicitly.
That is not the same thing as a logged-in person using the ChatGPT app. These are official interfaces with web search on, but they are measurement surfaces rather than consumer sessions: no personalisation, no memory, no app-only features. Results can differ from what any individual user sees, and we do not claim the two are identical. We would rather put that on the pricing page than in the small print.
How this page is sourced
- Vendor pricing and features come from each vendor's own public pricing page, with the date we checked it. We do not copy figures out of other people's comparison posts.
- Google-specific claims come from Google Search Central, not from SEO commentary about Google.
- Research claims name the paper and separate what was demonstrated in a benchmark from what holds in the open web.
- First-hand numbers come from our own measurement runs. Where the brand measured is founder-owned, the page says so in the same sentence as the finding.
This page is maintained by the person who built the measurement engine. There is no editorial team and no external reviewer behind it, and we are not going to invent one for a byline. How the measurements are collected.