How Laup measures whether AI names you.
You choose the product or solution categories and questions that matter. Laup asks each configured AI engine the same questions, stores the answers and provider-exposed evidence, then tracks who appears and how that changes over time.
One percentage, defined before it is displayed.
AI Visibility is not a model score. It is a count derived from captured answers, with the same denominator rules applied to you and every other company.
For each engine
AI Visibility
An engine enters the combined view once at least one answer in the selected window contains an attributable company or domain.
Answers that name or cite you
Answers with at least one attributable company or cited domain
Combined view
The equal mean of the configured engine percentages. Raw answer and appearance counts remain visible beside the percentage.
From a buying question to a trend.
The sequence is fixed. Capture comes before interpretation, and every derived measure keeps a path back to the answer it came from.
01
Question
The selected buying question, category and locale are snapshotted for the run.
02
Engine answer
Every configured engine receives the same question text and the matching locale signals.
03
Evidence
Laup stores the answer, citations and provider-exposed search or grounding context.
04
Appearances
Companies named in prose and domains cited in the answer are attributed and counted.
05
Trend
The same measures are compared by category, engine and trailing time window.
Laup proposes the starting set. You decide what runs.
Suggestions are tailored to your categories, market and language. Before tracking starts, you can edit the wording, add your own questions or remove any that do not fit.
Category questions
Within your plan total
Unbranded buying questions for the product or solution categories you choose. Laup can suggest up to 11 per category, across the categories your plan tracks.
“Best robot vacuum for pet hair in a small apartment?”
General market questions
Up to 8 suggested
Unbranded questions about the broader market you compete in. They show who owns the category beyond one product or solution set.
“Which online electronics stores are worth considering in Norway?”
Branded questions
Up to 3 suggested
Questions that name you directly. They support the sentiment view and stay outside the AI Visibility percentage, so direct questions cannot inflate it.
“Is this company reliable, and what do customers say about it?”
The selected text is snapshotted with every answer. Locale travels beside it as the market, language and provider-supported location signals. Laup adds no competitor names or merchant-specific persuasion to an unbranded question.
The same question, measured through distinct stacks.
These are controlled API measurements, not recordings of a logged-in consumer session. Each stack has its own model and search backend, which is why the answers can diverge.
ChatGPT view
OpenAI Responses API with OpenAI web search
The direct OpenAI measurement used for the ChatGPT view.
Copilot view
Azure OpenAI Responses API with Bing-grounded web search
A Bing-grounded GPT measurement used as the Copilot view.
Gemini view
Google GenAI with Google Search grounding
Google's model and search-grounding stack.
Claude view
Anthropic Messages API with native web search
Available where Claude coverage is included in the configuration.
Coverage depends on the plan and site configuration. Laup stores both the configured model and any model snapshot the provider returns, so a dated answer remains tied to the stack that produced it.
The answer, the exposed evidence and the change over time.
The final answer is the primary record. Search activity and citations add context, with clear boundaries around what each signal can establish.
Evidence captured
The answer
The final response text from each engine, stored with the configured model and provider-reported model snapshot.
Mentions and citations
Who is literally named in the prose, and which sources the final answer explicitly uses or points to.
Search and grounding context
Queries, retrieved pages and surfaced sources where the provider exposes them. Context does not by itself prove use or causality.
Reasoning summaries
Provider-exposed reasoning or thought summaries where available. Laup does not claim access to hidden reasoning a provider does not return.
Measures derived
AI Visibility
How often you appear in attributable answers, calculated per engine and then averaged equally across engines.
Average position
Your mean rank among the companies located in answers where you appear. Lower is better.
Grounding rate
The share of captured answers where the provider reports that the engine ran a web search.
Sentiment
Positive, neutral or negative treatment in answers that literally name a company, including the separate branded question set.
A citation shows that the final answer used or pointed to a source. A grounding or search source shows that the provider exposed it in the answer context. Either can inform investigation, but neither alone proves why the engine chose a company.
The same rules apply to you and everyone else.
Classification affects who belongs on the competitor list. It does not rewrite the captured answer or hide the source and appearance counts behind it.
One answer is one counting unit
A company appears once in an answer when its domain is cited or its name appears in prose. The citation and prose layers remain visible separately, even when both occur in the same answer.
Named means literally named
The extraction pass must return a supporting quote that exists in the answer text. Laup validates the quote before the mention is accepted.
The competitor lens follows your business
Retailers are compared with retail alternatives, product brands with manufacturers, and software companies with software alternatives. Customers can correct classifications and competitor relationships.
Engines keep equal weight
AI Visibility is calculated separately for each configured engine. The combined view is their plain mean, so an engine with more captured answers cannot dominate the result.
Competitors emerge from measured answers
There is no required list to maintain before tracking begins. Laup discovers recurring companies from the answers, classifies them, and preserves customer corrections as explicit overrides.
You receive no scoring preference
Your company goes through the same mention and citation pipeline as every other company. Deterministic matching identifies which attributed entity is you.
One category, two very different engine views.
Illustrative data for a retailer tracking one category. The calculation mirrors the live product; the companies and values do not represent a customer.
ChatGPT view
66 answersYou
46.2%
24 of 52 attributable answers
Category leader
55.8%
29 of 52 attributable answers
Copilot view
66 answersYou
18.6%
8 of 43 attributable answers
Category leader
41.9%
18 of 43 attributable answers
Combined AI Visibility
Your result is the equal mean of the two engine percentages: (46.2% + 18.6%) ÷ 2 = 32.4%. The two engines contribute equally even though their attributable-answer denominators differ.
Robot vacuums, trailing 30 days
132 captured answers
| Company | AI Visibility | Appeared | Cited | Named | Avg position |
|---|---|---|---|---|---|
| Retailer A | 48.8% | 47 | 31 | 35 | 1.8 |
| YouYou | 32.4% | 32 | 18 | 23 | 2.6 |
| Retailer B | 30.9% | 30 | 21 | 17 | 2.9 |
| Retailer C | 19.3% | 19 | 11 | 13 | 3.4 |
| Retailer D | 9.5% | 9 | 6 | 5 | 4.1 |
You appeared in 32 answers and ranked second in the combined category view at 32.4%. The category leader reached 48.8%. The gap is a measurement, not a diagnosis: the source, answer and Action views provide the evidence needed to investigate what may be behind it.
What the measurement can establish, and what it cannot.
Raw capture remains the source of truth
Laup stores the literal provider response where the API exposes it, and a lossless structured dump where it does not. Derived tables can be rebuilt from that capture.
Model context stays attached
The configured model, provider-returned model snapshot, prompt text, locale and run date stay with each response. A historical answer is never presented without its measurement context.
Failures remain visible
Retries and failed attempts are persisted. Rate-limit protection stops a run rather than silently filling a gap with invented or partial data.
Evidence is not causality
A citation confirms that the answer used or pointed to a source. A grounding source confirms provider-exposed context. Neither alone proves why the engine chose a company.
Before you rely on the number.
Do I choose the questions Laup tracks?
Yes. Laup proposes a tailored starting set from the product or solution categories, market and language you provide. You review the set before tracking starts, and can edit, add or remove questions within your plan coverage. The selected wording is snapshotted with every run so historical answers remain auditable.
Why does my AI Visibility differ between engines?
Each engine uses a different model and search stack, so identical question text can retrieve different sources and produce different answers. Laup keeps every engine view separate. The combined percentage is the equal mean of the configured engine percentages, so one engine cannot dominate because it returned more answers.
Does a citation prove why an engine recommended a company?
No. A citation shows that the final answer explicitly used or pointed to a source. A grounding or search source shows that the provider exposed it in the answer context. Either can be relevant evidence, but neither alone proves that the source caused a recommendation.
Why do some answers show no web sources?
The engine decides whether to search. It may answer from learned knowledge without exposing a web source. Laup still stores and analyses the final answer, including companies named in its prose, while clearly recording that no provider-reported web search occurred.
How often do the numbers update?
Current plans run daily. Each run asks the selected question set on every engine configured for the site. Dashboards use trailing windows, so the figures reflect repeated visibility rather than one isolated answer.
How we score AI readiness
AI readiness checks whether crawlers can access and interpret the site. It is useful context, not a prerequisite for measuring whether AI names you.
Read the readiness methodologyHow evidence becomes an Action
The Action plan investigates likely root causes, ranks the work and keeps the result connected to the evidence it came from.
Read the Action methodologySee where AI names you, and where it does not.
Start with your own questions and compare the engines your buyers use.
14 days. No credit card.