Prompt source and wording
Record why the question belongs in the panel and preserve the exact wording.
AI Visibility · Measurement
Measure AI answers without calling them revenue.
Record an AI answer as an observation, not a business result. Keep the exact prompt, surface, date and output. Then report platform data, visits, qualified actions and commercial outcomes as separate records.
Six separate records
Start with the source of the number.
Ask whether the number came from an official platform, an analytics tool, a synthetic prompt panel or the business’s operating system. Each source answers a different question.
Mindflow Marketing keeps these records separate before comparing dates. That prevents a sampled answer from becoming an unsupported claim about audience reach, leads or revenue.
| Layer | Record | What it establishes | What it does not establish |
|---|---|---|---|
| Answer observation | Exact retained output | What one logged run showed | Audience reach, permanence or buyer influence |
| Official platform record | Impression, citation, page or query field | What that platform reported | The complete conversation or buyer choice |
| Visit | Referral session and landing page | A trackable click path reached the site | Qualification or a sale |
| Onsite action | Defined analytics event | An instrumented action occurred | Lead quality or commercial value |
| Qualified action | Approved operating criteria | The action met a business rule | A completed commercial result |
| Commercial outcome | Scheduling, CRM or commerce record | The business recorded the result | Sole causation by an AI answer |
Observation ledger
Log the conditions before you score the answer.
AI outputs can change with the prompt, surface, interface mode, model, account, location, language and time. A screenshot without those conditions is hard to compare and easy to overstate.
Use one observation ID per run. Keep the full output, not only the sentence that mentions the business. Record the method version and retest date so later checks use the same disclosed rules.
Engine, interface and mode
Name the product and the mode used. Do not combine results from different surfaces.
Session and location
Record account state, geography, language and device where they are known.
Timestamp and version
Capture the date and any exposed model or interface version.
Full answer evidence
Retain the complete response, visible sources, cited URLs and answer position.
Method and next check
State the calculation, panel version, notes and conditions for the next run.
Different answer signals
Keep mentions, recommendations and citations separate.
A mention means the answer named the business. A recommendation means the answer presented it as an option. A citation means the answer exposed a source. One run can contain any combination of the three.
Position, factual accuracy and sentiment are separate fields. A prominent inaccurate mention is not better evidence than a lower accurate one. A citation may support a claim without naming the business at all.
The business was named
Useful for presence in the sampled output. It does not establish endorsement, source use or a visit.
The business was framed as an option
Keep the exact wording and qualification. It is not a stable rank or a future promise.
A source was exposed
Record the cited page and relationship. The citation does not prove a click or buyer trust.
The facts were checked
Compare material names, services, products, people, locations and limits with current sources.
Show the denominator
Put the calculation beside every score.
A count can be honest. A percentage can be honest. A share can be useful. The problem begins when the report hides the denominator, prompt source, competitor set or weighting rule.
“7 of 12 logged tests” describes a disclosed sample. It is not audience share, market share or the probability that a buyer will see the business. Repeated runs can reveal variation, but no universal run count creates certainty.
Mindflow Marketing shows the denominator beside the sampled score and keeps the commercial result elsewhere.
| Metric | Required denominator or method | Safe interpretation | Do not call it |
|---|---|---|---|
| Mention count | Logged runs in the disclosed panel | Presence within that sample | Total audience reach |
| Recommendation rate | Recommendation runs ÷ eligible logged runs | Sample frequency under those conditions | Conversion probability |
| Citation share | Citations, source set and comparison scope | Relative source appearance in the sample | Universal authority |
| Proxy impressions | Vendor model and external search-volume inputs | A constructed estimate | Actual platform impressions |
| Competitor share | Named competitor set and weighting | Comparison inside the selected panel | Whole-market share |
Keep X/12 as a disclosed sample—not the headline business result.
First-party platform records
Use official data where it exists.
Official platform reporting narrows uncertainty because the platform defines and exposes the field. It still has scope, availability and aggregation limits.
Third-party tools remain useful for controlled prompt panels and comparisons. They do not have hidden access to a platform’s internal ranking or AI systems.
Mindflow Marketing prefers the official field for the named platform, then uses synthetic panels to answer only the questions the platform does not report.
Generative-AI impressions
Google’s current report can expose impressions by page, country, device and date for eligible properties. Availability and data remain limited; it does not reveal every prompt or lead.
Citations and cited pages
Bing reports citation activity and sample grounding queries. A citation count does not show answer placement or page importance.
Citations and AI referrals
Clarity exposes citation activity, cited pages, authority share and AI-referred session percentage as separate fields. It does not turn them into one causal score.
Trackable referral clicks
OpenAI documents a source parameter on ChatGPT search referrals. That can identify a click in analytics when the visit is recorded; it does not reconstruct no-click influence.
Events and attributed credit
An onsite action becomes a key event when the business defines and instruments it. Attribution assigns credit under a declared model; it is not the same as causation.
Platform sources checked 16 August 2026. AI reporting fields are changing quickly and require a current pre-publication review.
After the answer
A visit is still not a qualified outcome.
A referral session says the site received a trackable visit. An onsite event says the instrumented action occurred. Neither tells the business whether the request fit, whether the appointment was appropriate or whether an order remained profitable.
Connect the visit to the approved operating record. Keep analytics attribution, human qualification and the commercial result as separate conclusions.
Referral session
Record source, landing page, date and available campaign fields.
Onsite action
Record the call click, form, booking request, demo, trial, cart or checkout event.
Qualification
Apply the business’s service, access, matter, buyer or order criteria.
Commercial record
Confirm the result in field service, scheduling, intake, CRM or commerce systems.
Five outcome records
Change the commercial record by industry.
The observation method can stay consistent. The qualified action, privacy boundary and source of commercial truth change.
| Industry | Qualified action | Commercial outcome | Required boundary |
|---|---|---|---|
| Home Services | Valid service request | Booked or completed job | Match call or form records to the approved field-service record |
| Healthcare | Appropriate appointment request | Scheduled appropriate appointment | Use privacy-aware collection and an approved scheduling record |
| Legal | Prospective-client consultation request | Qualified or signed matter where approved | Use reviewed intake, jurisdiction and confidentiality controls |
| B2B SaaS | Demo or trial action meeting criteria | Qualified opportunity or pipeline | Align product analytics and CRM definitions |
| eCommerce | Product, cart or checkout action | Order, return and repeat purchase | Use the commerce platform as the commercial record |
Claim-confidence key
Label what the evidence supports.
Every conclusion should say whether it was observed, reported, inferred, attributed or remains unknown. The label shows how far the evidence can travel.
| Label | Use it when | Example | Boundary |
|---|---|---|---|
| Observed | Mindflow retained the dated answer | The business appeared in 7 of 12 logged tests | The finding belongs to that sample |
| Reported | An official platform exposed the field | Search Console reported a stated number of generative-AI impressions | The platform definition controls |
| Inferred | Records moved together but cause is unresolved | Branded visits rose during the same period | Timing alone does not prove cause |
| Attributed | A declared model assigned credit | Analytics assigned last-click credit to a ChatGPT referral | Name the attribution model |
| Unknown | Evidence is missing or incompatible | No-click influence cannot be reconstructed | Do not fill the gap with a score |
Decision-ready reporting
End every report with the next decision.
A useful report says what changed, how confident the team is, who owns the action and when the evidence will be checked again. It also says what new information would reverse the decision.
Mindflow Marketing uses the record to choose one priority: correct an inaccurate source, improve a page, repair access, change a test panel, fix the handoff or leave the system alone until more evidence arrives.
Keep the raw evidence beside the summary. A decision-maker should be able to move from the headline number to the exact prompt run, platform field, referral record or business definition that produced it. If two systems use different date ranges or definitions, show the mismatch instead of forcing a comparison.
Record a “no action” decision when the movement is small, unstable or outside the business’s control. Waiting for a better sample can be more responsible than changing a useful page after one noisy check.
Name the movement
State the source, period, baseline and exact field that changed.
Name the evidence label
Use observed, reported, inferred, attributed or unknown.
Name the owner and due date
Assign the smallest consequential next step.
Freeze the next conditions
State when, where and how the comparison will run.
Say what could change the decision
Define the missing evidence, threshold or operating fact that would lead elsewhere.
Review the evidence ladder
Know which numbers are official, sampled or commercial.
Bring one AI-visibility report or baseline. Mindflow Marketing can show which numbers are official platform fields, synthetic observations, behavioural records, qualified actions or commercial outcomes—and which claims the evidence cannot support.
The review does not promise citations, score gains, leads or revenue attribution. Healthcare and Legal applications remain behind their recorded privacy, clinical, jurisdiction and intake review gates.
Primary platform and measurement sources checked 16 August 2026. Re-check within 30 days of publication.