Exact business
The selected location, website, name, and category should agree. A national brand cannot silently replace its local branch.
Owner guide · August 19, 2026
What should an AI visibility audit actually prove?
A useful audit should do more than produce one impressive-looking score. It should show where the business appears, who appears instead, what evidence supports the answer, and which next move is worth making.
1confirmed business entity
1locked market or location
4separate decision measures
Datedbaseline for honest retesting
The plain answer
An AI visibility audit should tell an owner whether the right business is being understood and recommended for the right questions in the right market. It should also reveal the competing businesses and source gaps behind that result without pretending a live answer is permanent.
01 · Start with identity
If the business or market is wrong, every number after it is noise.
Chains, similar names, multiple locations, and broad categories can produce convincing but irrelevant results. A serious audit confirms the entity, canonical website, location, category, and comparison market before scoring anything.
Exact market
Franklin business law should be compared with similar firms serving Franklin, not unrelated lawyers or a generic national list.
Exact questions
The prompts should reflect real buyer decisions and remain unchanged when the audit is repeated.
02 · Keep the measures separate
Four numbers answer four different questions.
Blending every signal into one score hides the useful part. Owners should be able to see what improved and what did not.
01
AI mentions
How many tested questions clearly named or recommended the business? A citation is not automatically a recommendation.
02
Recommendation position
Where did the business sit among comparable choices in the locked market? Lower positions need competitor context.
03
Website readiness
Can crawlers reach and understand services, locations, credentials, proof, structured data, and the next action?
04
Confidence and source status
Did the live provider answer? Was a fallback used? Was the entity match strong enough to support the conclusion?
03 · Example report shape
A result should be readable in one minute.
The example below is fictional. It shows how the measures should be presented, not a client result or benchmark.
| Measure | Illustrative result | What the owner learns |
|---|---|---|
| AI mentions | 2 of 5 questions | The business is understood, but not consistently selected. |
| Recommendation position | 7 of 12 | Six comparable businesses were surfaced earlier. |
| Website readiness | 68/100 | Core information is readable; proof and service depth need work. |
| Confidence | Medium-high | The entity matched, but one provider response was incomplete. |
A provider failure should be shown as unavailable, not converted into a zero. A zero is an observed result; unavailable means the test did not return enough evidence to score.
04 · Connect visibility to the business
The score is not revenue. The path from discovery to revenue can be measured.
The owner needs a simple chain: qualified visibility, website visit, call or form, booking, and customer value. Visibility only creates the opportunity to enter that chain.
Illustrative economics
If improved discovery produces 100 additional qualified visits, 6% become enquiries, 40% close, and the average first purchase is $600, the scenario produces about $1,440 in first-purchase revenue. That is a planning model, not a forecast. Replace every input with the business’s actual analytics and sales data.
Discovery
Track tested mentions, supporting citations, search impressions, qualified visits, and the landing pages receiving them.
Conversion
Track calls, forms, bookings, conversion rate, lead quality, and the questions that preceded the visit.
Economics
Track close rate, customer value, gross margin, capacity, and payback—not a visibility score in isolation.
05 · Limits
A credible audit says what it cannot prove.
AI results can change with wording, date, location, source availability, model updates, and personalization. Technical eligibility also does not guarantee inclusion.
No permanent rank
A dated answer is a reproducible observation, not a permanent position for every buyer or every engine.
No secret algorithm
Public evidence can be audited. A platform’s private weighting cannot be reverse-engineered with certainty.
No automatic ROI
Better visibility may create more qualified opportunities. The offer, website, follow-up, capacity, and sales process still determine revenue.
06 · Owner checklist
Seven questions to ask before trusting the report.
If the report cannot answer these plainly, the score is probably doing too much work.
01
Was the exact entity confirmed?
Name, location, category, and canonical website should agree.
02
Was the market locked?
The comparison set should match the buyer’s geography and service need.
03
Can I see the questions?
The prompt set should be relevant, dated, and reusable.
04
Are mentions and rank separate?
Being named once is not the same as being the leading recommendation.
05
Are provider failures disclosed?
Unavailable evidence should never be disguised as a poor score.
06
Can I trace the source?
Competitors, citations, site evidence, and profile evidence should be reviewable.
07
Is there a retest plan?
Improvements should be judged against the same baseline and business outcomes.
Primary references
What the platforms say.
Google says the same foundational SEO practices apply to AI Overviews and AI Mode: allow crawling, use clear internal links, keep important information in text, and make structured data match visible content. OpenAI provides separate controls for search discovery and model training.
See the evidence behind your visibility.
Start with a market-locked preview, then decide whether the gap is worth fixing.