Knowledge

Research

Here we publish the results of our own measurements of brand visibility in AI responses. The first study covers a single brand, a fixed set of 18 queries, and measurements from June, July and September 2026. Key finding: the brand went from zero mentions to being the most frequently cited source in its category, yet the models still do not recommend it.

Last updated: 14 September 2026 · reading time: approximately 7 minutes

Method

What exactly did we measure?

The study covered a single brand: a regional manufacturer of building materials and fertilisers, selling mainly within a single province. A set of 18 purchase enquiries was fixed prior to the first measurement and remained unchanged between scans.

Between the first and last measurements, the set of engines and the method used to calculate the average visibility index changed. We have therefore not included visibility figures as percentages here: the values from June and September are not comparable. We have retained only those indicators which we have calculated in the same way throughout.

All three are simple counts: how many queries mention the brand, in how many models do they refer to its domain, and how many search engines recommend the brand without any prompts. We describe the counting method in methodology.

Study parameters

BrandOne. A regional producer of building materials and fertilisers, operating within a single province.

Enquiries18, frozen prior to the first measurement, identical in every scan.

MeasurementsJune, July and September 2026. In the intervening period, the brand worked on raising its profile.

EnginesFour in the latest survey. The sample changed over the course of the survey, which limited the comparability of some of the indicators.

We do not mention any brand names. The study describes the mechanism, not a specific company.

Results

What did the subsequent measurements reveal?

Two curves are rising together. The third one hasn’t budged.

Study procedure; measurements taken in June, July and September 2026

Search queries mentioning the brand Search queries containing the brand’s domain name

Both figures are calculated using the same set of 18 queries in each measurement. The vertical axis shows the proportion of queries, not the average visibility metric.

Brand mentions0 out of 18 → 17 out of 18 → 18 out of 18In June, the brand did not appear even once, although the models discussed its category at length. In September, it appears in every query in the set.

Citing your own domain6% → 50% → 94%In the latest survey, the brand’s domain is the source for 17 out of 18 queries and is cited more often than any other domain in the category.

Recommendation0 → 0 → 1 engineThe findings have remained the same for three months: consumers are aware of the brand but do not recommend it of their own accord. The barrier described is identical in every report.

At that time, the brand was working on its own website: it was updating product pages, adding information about the company and organising content to address buyers’ queries. It did not receive any independent reviews or achieve any rankings during this period. The first two metrics reflected this work, but the third did not.

Conclusion

Visibility and a recommendation are two different things

Working on their own website was enough for the models to start mentioning and quoting the brand. It wasn’t enough for them to start recommending it.

The fact that there are no mentions does not mean the topic is dead In the first survey, the models discussed the category at length and named specific suppliers. The brand simply wasn’t mentioned. The absence of any mention is an indication of the brand, not of a lack of interest in the topic.

A custom domain triggers two out of three indicators Organising the content and data describing the company resulted in mentions and citations within a single month. This is the fastest traffic available.

Being the most frequently cited source does not constitute a recommendation In the latest survey, models refer to the brand’s website more often than to any other source in the category, yet they still do not recommend it. The fact that it is cited is evidence of the content’s availability, not of its quality.

The barrier remained the same for three months In every report, the models highlighted the same shortcoming: a lack of independent rankings, comparative tests and external reviews. No changes made to their own website have altered this.

We describe this condition separately in the article knows it, but doesn’t recommend it. The study shows how much work can be done in it without changing it in the slightest.

Observation

The same brand, the same questions, different answers

Data from a single measurement, taken in September 2026. Four engines received an identical set of queries on the same day.

Proportion of responses mentioning the brand, measured in September 2026

Claude

100%average position 1.8

Perplexity

100%average position 11.2

ChatGPT

56%average position 10.3

Gemini

44%average position 12.9

Mentions and rankings are two separate things. Both search engines mention the brand in every result, but one places it in the top two, whilst the other ranks it only around eleventh place.

A practical point concerns how to check this. If someone types a query about their own category into a chat and sees their own brand in first place, they are looking at the result from a single search engine, not an accurate picture of their visibility. The same brand, in a different search query on the same day, appears in less than half of the results. We explain how to check this for yourself in the article how to find out what AI says about a company.

To be honest

What this study does not show

Limitations you need to be aware of before applying these figures to your own category.

It’s one brand, not a trialThe results describe a specific case within a specific category. They do not provide a basis for making claims about the market or about how another sector might behave.

This category is specificA technical product, regional sales, competition comprising several large manufacturers. In a saturated market, the pace would be different.

The method changed during the course ofBetween measurements, the set of engines and the method used to calculate average visibility changed. We are therefore only showing the figures that have been calculated in the same way from the start.

This is not proof of causationWe can see that the metrics have increased following work on our own domain. As we do not have a control group, we cannot rule out the possibility that part of the change is due to updates to the models themselves.

Future studies will cover more brands and categories. Until then, please treat these figures as a single case study rather than the norm.

FAQ

Questions about this survey

Questions about the methodology, the data and what this implies for other categories.

How many brands were included in this survey?

Just one. This is a single-case study conducted over three months using a fixed set of 18 queries, not a sample-based study. The results describe a mechanism, not a market distribution.

Why don’t you give the visibility index as a percentage?

This is because, between the measurements, the set of engines and the method used to calculate the average changed. The figures from June and September do not represent the same quantity, so plotting them on the same graph would be misleading.

Did the brand pay to have these results published?

No. These are data from measurements carried out as part of routine maintenance, published without the company’s name or the names of its products.

How do we know that the increase was due to work on the website?

It is not known for certain. We can see a temporal correlation, but without a control group, it is impossible to distinguish the effect of the work from the updates to the models themselves. We therefore describe this as an observation, not as proof of causation.

Will it be the same in my sector?

The pace depends mainly on how saturated the category is. The brand under review operates in a regional niche, where the models had few specific indications. In a category with dozens of recognisable players, the first mentions tend to come later.

What needs to be done to ensure that the model starts recommending, rather than just listing?

Obtain independent quality endorsements from outside your own sphere: reviews, comparative tests, rankings. We discuss this in the article knows it, but doesn’t recommend it.

How often did you repeat the measurement?

Once a month, using the same set of queries. We explain why this is the case in the article How often should visibility be measured?.

Will there be further studies?

Yes, as we gather data from more brands and categories. Further publications will appear on this page.

See where your brand stands

You start with 250 AeloCoins, without a card. These are enough to start a measurement and test the modules. Once you’ve used them up, you switch to a paid plan. Measurements start straight away, and a full report on your brand is included from the Starter plan onwards. You’ll see mentions, citations and brand awareness figures separately, just as in this study.

Create an account

Research

What exactly did we measure?

What did the subsequent measurements reveal?

Visibility and a recommendation are two different things

The same brand, the same questions, different answers

What this study does not show

Questions about this survey

See where your brand stands

Product

Knowledge

Company

Research

What exactly did we measure?

What did the subsequent measurements reveal?

Visibility and a recommendation are two different things

The same brand, the same questions, different answers

What this study does not show

Questions about this survey

See where your brand stands

Product

Knowledge

Company

Sign up for the start Aelo