An AEO dashboard should show the chain from technical eligibility to commercial outcome without pretending every link in that chain is observable. You need to know what appeared, where it appeared, which source was shown, whether anyone visited and what happened next.
If the report compresses that into one “AI visibility score”, it may be easy to present and impossible to act on.
The seven metrics that matter
| Metric | What it tells you | Decision it supports |
|---|---|---|
| Technical eligibility | Whether priority pages can be crawled, indexed, rendered and shown | Fix access before optimizing the answer |
| Query-set coverage | Whether each important question has one accountable page | Create, improve or consolidate the right route |
| Answer presence | Whether the brand appears in a repeatable observation | Diagnose omission, misclassification or volatility |
| Citation and source presence | Which URL or third-party source is exposed with the answer | Strengthen, correct or replace the source |
| Search and referral visits | Whether observable discovery generated a visit | Improve the destination and acquisition path |
| Conversion contribution | Whether the journey contributed to a qualified action | Prioritize pages and queries with commercial value |
| Shipped changes | What changed, when and why | Connect movement to interventions without inventing causality |
Every metric should be filterable by platform, query cohort, market, page, date and business priority where the data allows.
Start with evidence states, not charts
A decision-grade dashboard labels every number as one of four states:
- Observed: captured directly from the platform, page or analytics event.
- Derived: calculated from observed inputs with a visible formula.
- Inferred: a plausible interpretation that has not been directly observed.
- Unavailable: the platform or account does not expose the data.
That last state matters. Unavailable does not mean zero.
If a prompt-monitoring tool did not capture a mention, it may mean the answer did not appear in that controlled run. It does not prove no customer ever saw it. If an analytics platform recorded no ChatGPT referral, it does not rule out a customer seeing an answer and returning later through branded search.
1. Technical eligibility
Before tracking prompts, prove that the source can enter the system.
For each priority URL, report:
- crawl status for the relevant bot;
- indexability and canonical target;
- sitemap membership where intended;
- rendered status and important textual content;
- internal links from the commercial parent and related sources;
- structured data validity and visible-content match;
- mobile and desktop defects that block comprehension or action.
Google says a page must be indexed and eligible to appear in Search with a snippet to be eligible as a supporting link in AI Overviews or AI Mode. It also says eligibility does not guarantee selection. See AI features and your website.
OpenAI says OAI-SearchBot access affects whether page content can be included in ChatGPT search summaries and snippets. Report crawler policies separately; one allow or disallow rule does not describe every platform.
2. Query-set coverage
Your dashboard needs a fixed commercial question set, not a random sample that changes every month.
Group it by:
- category and service discovery;
- problem and solution research;
- comparison and alternatives;
- pricing, scope and implementation;
- trust, proof and risk;
- location or market where relevant;
- branded fact accuracy.
Then assign one page to each distinct intent. Report gaps, duplicate ownership and orphan routes.
This metric prevents a common failure: celebrating more content while several articles compete for the same query and the actual service page remains weak.
3. Answer presence
For each controlled observation, retain:
- exact query or prompt;
- platform and interface;
- market or location;
- device where available;
- account or personalization state where known;
- retrieval date and time;
- answer text or evidence capture;
- whether the brand appeared;
- how the brand was described;
- confidence and limitations.
Do not report a single run as a stable market share. Generative answers can vary. Use repeated observations and show the denominator.
A useful metric is:
answer presence rate = observations containing the brand ÷ valid observations in the fixed cohort
The formula is simple. The discipline is keeping the cohort stable enough for the trend to mean something.
4. Mentions, citations, links and sources
These are different events:
- a mention names the brand;
- a citation attributes part of the answer to a source;
- a link creates a destination;
- a supporting source is a page exposed by the interface;
- a referral visit reaches your analytics.
Track them separately.
For each citation or source, record:
- owning domain and URL;
- whether it is your page or an independent source;
- the claim the answer appears to associate with it;
- whether the visible source actually supports that claim;
- destination status and canonical target;
- source quality and conflict risk.
A citation does not prove the source caused the entire answer. It gives you a visible source relationship to audit.
5. Google Search visibility
Keep normal Search Console reporting:
- impressions;
- clicks;
- click-through rate;
- average position;
- query;
- page;
- country;
- device;
- date.
Google announced dedicated generative-AI performance reports in Search Console in June 2026. The initial rollout is limited to a subset of sites and can include impressions, pages, countries, devices and dates. Check the account before promising this view. See Google's announcement.
Where the dedicated view is unavailable, report it as unavailable. Do not manufacture an “AI Overview click” breakout from aggregate web data.
6. AI referral traffic
Report observable AI referrals by:
- source and medium;
- landing page;
- session quality;
- conversion event;
- assisted journey;
- new versus returning user where appropriate.
OpenAI says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs from ChatGPT search, allowing tagged inbound traffic to be analyzed. That is useful observable evidence. It is still only the click path.
Keep direct, branded-search and untagged journeys separate. Do not quietly relabel them as AI traffic.
7. Commercial outcomes
The commercial layer should answer:
- Which priority pages generated qualified enquiries or sales?
- Which answer and search cohorts contributed to those journeys?
- What was the landing-page conversion rate?
- Which leads were accepted, progressed or closed?
- What revenue can be attributed under the agreed model?
- Where is attribution unavailable?
Use the business's real conversion events, CRM states and value definitions. A mention is not a lead. A citation is not revenue. A visit can contribute to a sale without deserving all the credit.
What should the executive view show?
The front page can be small.
Outcome
- qualified conversions from organic and observable AI referrals;
- pipeline or revenue where attribution is supported;
- material changes against the prior comparable period.
Visibility
- search demand and page performance;
- answer presence by fixed cohort;
- citation and source mix;
- wrong or missing brand facts.
System health
- blocked or non-indexable priority pages;
- duplicate intent ownership;
- source and evidence gaps;
- conversion-path defects.
Next action
- the single highest-value change;
- owner;
- dependency;
- expected signal;
- retest date.
The dashboard is there to force a decision, not to display every available metric.
What should the working view show?
The delivery team needs more detail:
| View | Required fields |
|---|---|
| Query cohort | Query, intent, market, priority, owner URL and platform |
| Observation log | Date, answer, brand state, citation, link, source and capture |
| Page ledger | Indexability, canonical, template, internal links, structured data and conversion event |
| Source ledger | Claim, source owner, publication date, support status and conflict |
| Change ledger | URL, change, release evidence, hypothesis and retest |
| Commercial ledger | Landing page, event, lead state, value and attribution boundary |
This is the measurement layer of the Managed Search Loop: observe the market, ship a targeted improvement, retest the same evidence and choose the next move.
Red flags in an AEO report
- A proprietary score has no visible denominator.
- “Share of voice” mixes markets, prompts and platforms.
- A mention, citation and link are treated as the same event.
- Missing platform data is displayed as zero.
- The query cohort changes between periods.
- Screenshots are missing dates, prompts or platform context.
- Traffic is attributed to AI without an observable referral or agreed model.
- Content published is presented as an outcome.
- No report shows which page or source should change next.
How often should you review it?
Use a cadence that matches the decision.
- Technical release: check immediately after release and after recrawl.
- High-value page or source change: retest the fixed cohort on the agreed schedule.
- Executive commercial view: monthly is often enough.
- Strategic reforecast: quarterly or when the market, offer or measurement model changes materially.
More frequent checks do not create more reliable conclusions when the denominator is tiny.
Frequently asked questions
What is the most important AEO metric?
There is no single metric. The useful chain is eligibility, answer observation, source attribution, visit and commercial outcome. The current bottleneck determines which metric matters next.
Are rankings still relevant?
Yes. Search demand, rankings, impressions and clicks remain valuable. AEO adds answer and source evidence; it does not remove SEO reporting.
Can a dashboard show every AI answer mentioning my brand?
No. A controlled tool can show what it observed for its query set and conditions. It cannot prove exhaustive visibility across every user, interface and personalized answer.
Should an AEO dashboard include structured data?
It should report validity and visible-content match where structured data is relevant. Schema is a technical input, not an outcome.
How do I forecast value from these metrics?
Use bounded scenarios and keep click-led, assisted and risk-reduction paths separate. See AEO forecasting.
How does this relate to broader AI visibility tracking?
For the observation process behind these metrics, use the AI search visibility tracking guide.
Make every metric earn a decision
Your AEO dashboard should tell you which question, page or source changed; what evidence supports that conclusion; and what happens next. Anything else is reporting theatre.
We connect that instrumentation to the answer engine optimization service and the website improvements it is meant to guide. If your current reporting cannot find the constraint, show us the market.