Methodology
How Aelo measures brand visibility in AI
Aelo measures brand visibility in AI a neutral scan: it asks AI chatbots questions from your category without mentioning the brand name, and checks whether the brand appears in the response, in what position, alongside which other brands, and which sources the model uses. The questions come from frozen query set, so each measurement measures the same thing. Full scan It covers four models (ChatGPT, Claude, Gemini and Perplexity) and takes place once a month. We break down the results into five metrics: mention, position, proportion of responses, citation and sentiment. We measure attendance at X using a separate module.
Author: Patryk Pietrzyk · Last updated: 25 October 2026 · reading time: approximately 7 minutes
Measurement principle
What is a neutral scan, and why don’t we mention the brand name?
A neutral scan is a test in which we ask questions from your category without mentioning any brand names, and check whether the model mentions the brand of its own accord. If we were to ask the model directly about the brand, it would almost always give an answer, and the result would be inflated.
There is one exception, and we address it directly. Alongside spontaneous awareness, we measure comparative awareness, in which we deliberately mention the brand name in order to compare it with a specific competitor.
We report both metrics separately and never combine them into a single figure. Spontaneous awareness indicates whether the brand features at all in the responses. Comparative positioning shows how the brand fares when the customer is already familiar with both names.
Spontaneous visibility
no brand name in the enquiry
Enquiry
Which companies in Poland are involved in [category]?
Model response
- 1Brand A
- 2Your brandreference
- 3Brand B
The model must mention the brand themselves. We count the mention, the position and who is standing next to them.
Comparative position
name given deliberately
Enquiry
What [Your brand] differs from [competitor]?
Model response
Your brand
in view of
competitor
The result shows how the brand performs in a direct comparison, rather than whether it is visible.
Scan range
Which AI models are covered by the assessment?
ChatGPT, Claude, Gemini and Perplexity, in a single run, with results shown separately for each model and in aggregate. Each model responds differently, selects brands differently and uses sources differently, so one is not enough to assess visibility. We show the results separately for each model and collectively.
ChatGPTIt accesses the internet in real time and gives significant weight to fresh sources.
ClaudeHe combines his training knowledge with online research, and is cautious in his recommendations.
GeminiIt combines the model’s knowledge with Google search results.
PerplexityHe almost always cites his sources, so he’s best at showing where he’s quoted from.
Module XWe measure the brand’s presence on X separately, outside the scan across the four models. A call there costs many times more than on the other engines, and the responses follow a different logic, so mixing this with the rest would skew the overall result.
We update our set of models as the market evolves. We record any changes to the set in the report, as this affects the comparability of results over time.
First scan
A set of queries is generated from the topics in the brand card and is frozen
Another scan
The same set, the same models
Another scan
The same set, the same models
You’re changing the topics in the brand tab
Scan after the change
The set is being built from scratch; the report notes this
Illustrative figures. The drop from 29 to 12 queries looks like a regression on the graph, but it is actually a change in the unit of measurement.
Repeatability
Where do the enquiries come from, and why has the order been put on hold?
Enquiries arise from the topics described in brand card: what you’re selling, to whom and where. After the first scan, the setup is locked so that subsequent measurements take the same data.
If you change the topics in the brand tab, the set will change accordingly. The report takes this into account, because without this information, the drop in visibility would appear to be a decline, when in fact it is a change in the metric.
We discuss what happens between scans in more detail in the article Why aren’t the changes visible straight away?.
Records
What metrics does Aelo measure?
Five: mentions, ranking, share of responses, citations and sentiment. We do not reduce visibility to a single metric, as a single figure can be misleading. A brand may be mentioned frequently but never cited, or cited and poorly described.
01
Mentions
3 of 18
Do models switch brands in your set of queries, and if so, how often?.
02
Item
2 in reply
Where in the list does the brand appear? The higher up it is, the greater the chance that the customer will choose it.
03
Proportion of responses
12%
What proportion of all the mentions relating to the same set are yours?.
04
Citations
6%
What sources do the models use, and is your domain among them?.
05
Sentiment
neutral
Is the brand described in positive, neutral or negative terms?.
Sample data. We report each metric separately for each model and in aggregate.
The two figures that raise the most questions
How do we calculate the response rate and the citation index?
Proportion of responsesWe take your set of queries, count the mentions of your brand and your specified competitors in the models’ responses, and check what proportion of all mentions belongs to you. The conditions are identical for every brand, as the questions and models are the same, so the comparison is fair. We track changes in market share over time, because A trend tells us more than a single measurement.mentions of your brand ÷ mentions of all the brands being compared
Citation rateWe collect the sources that the models cite in their responses and group them by query and model. You can then see which domains dominate in your category and whether your website appears as a source at all. This is separate from mentions: a brand may be mentioned by name, yet never actually cited.replies mentioning your domain ÷ total number of replies
Measurement frequency
How often does Aelo carry out a full scan?
Once a month, and we record each result in the history. A single measurement is just a snapshot of the situation at a given moment. It is the history that is meaningful, which is why we show the trend.
Day 1Full scan
Day 30Another full scan
You implement tasks · the models process the changes
What does a full scan provide?
A more in-depth analysis of citations, sources, barriers and competition. It is from this that report and task list.
Why not more often?
Changes on the website and signals from outside it take time before the models pick them up.
What would a more frequent scan measure?
Variations in the models, not the result of your work. A single result should still be interpreted as a point within a trend.
We explain in the article how long the models take, why the changes aren’t visible straight away, and how the frequency is selected how often to measure visibility.
To be honest
What does measuring visibility in AI not tell us?
It does not say exactly how the model will respond to the next question, it does not guarantee growth, and it does not say whether the brand is recommended, only whether it is mentioned. We distinguish between two things: whether we diagnose the situation accurately and whether we change the outcome. We verified the diagnosis using the measurements described in research. We haven’t proven its effectiveness, and we’re not going to pretend we have.
The models are not deterministicThe same query can produce a slightly different result, and Models change with each update. That is why we treat a single measurement with caution and look at the trend.
We do not control the responsesNo one controls what the model generates. We measure the change in the number of mentions between a scan taken before implementation and a scan taken after it, with no promised repayment date and no interest.
Visibility is not a recommendationA brand can be recognised, yet not recommended. Diagnosis is being built content on your own domain, a recommendation based on signals from outside it, and it takes longer.
In a nutshell
Key points about the method
- Neutral scan: we ask the same questions as the customer, without mentioning the brand name.
- Four models, with results shown separately for each one and in total. Attendance at X is measured by a separate module.
- A frozen set of queries, with every change to it recorded in the report.
- Metrics: mentions, ranking, share of responses, citations and sentiment.
- A full scan once a month. It’s the trend that counts, not a single reading.
FAQ
Questions about the method
The questions we’re asked most often when discussing measurements.
Does Aelo provide the models with the name of my brand?
Not when measuring spontaneous awareness. The questions relate to the category, and the respondent must name the brand themselves. We deliberately only mention the brand name in the comparative section, which we report separately.
What will happen if I change the themes in the brand tab?
The set of queries will change along with them, and the report will reflect this. Comparability with previous scans is then lost, as it is the metric that changes, not the result.
What is the difference between a mention and a citation?
A mention is when a brand is named in the body of a reply. A citation is when your domain is cited as a source. A brand is sometimes mentioned but never cited. We collect these terms in dictionary.
Why can results vary?
This is because the models are not deterministic: the same query may yield a slightly different result, and the models themselves change with each update. That is why it is the trend across multiple scans that matters, not a single result. Read more in the article Why aren’t the changes visible straight away?.
Can I check for myself what the AI chatbot is saying about my brand?
Yes, but a manual test only shows the result for a single model at any one time. The same brand may appear in every response in one chat, yet in fewer than half the responses in another; that is why the scan covers four models and a fixed set of queries. We explain how to do this properly in manual test.
Does the measurement method depend on the plan?
No. The method is the same for every plan. The full report is already included in the Free plan, and the prices of the plans are in pricing.
What sources is this text based on?
- Aggarwal et al. (Princeton University, IIT Delhi), GEO: Generative Engine Optimisation: visibility metrics in generative responses.
- SparkToro and Gumshoe.ai: the consistency of brand recommendations in ChatGPT, Claude and Google AI: The attendance rate based on multiple responses is more reliable than the ranking (January 2026).
- OpenAI: bots and their user agents: OAI-SearchBot fetches web pages for use in ChatGPT’s search function.
- Google: Grounding with Google Search (Gemini API): Gemini combines the answer with the current search results.
- Anthropic: web search in Claude: answers with source citations (March 2025).
- Perplexity: how to get started: real-time search results, with sources.
- Aelo surveys: a single-brand survey conducted in June, July and September 2026 (described in research).
Measure your visibility in AI
Free Account No card required: you get the following to start with 250 AeloCoins and a full report on your brand’s visibility on ChatGPT, Claude, Gemini and Perplexity. The methodology is the same for every plan.
