The lime manufacturer’s third measurement, taken on 10 September 2026, revealed the brand’s first recommendation in AI: one of the four models began recommending the company, although none had done so in June or July. The website’s citation index rose from 6% to 94% over 11 weeks, and the brand has more mentions than any other manufacturer of its size.
In the third test, on 10 September 2026, a family-run lime producer from the Łódź Voivodeship appeared in all 18 queries in the set, and the AI models cited its website as a source for 17 of them (94%). All four models tested recognise the brand, and one has recommended it for the first time – marking the first instance of a brand recommendation by AI. This occurred 11 weeks after the first measurement and following two scans in which the brand was recognised but not recommended by any of the models.
This is the second part of the case study. The first part, From zero to 63% visibility in AI in five weeks, describes the measurements taken on 24 June and 27 July 2026, as well as the changes to the website that occurred in the intervening period. It concluded that all metrics were rising except for one: the recommendation. In September, that metric showed a change.
The third measurement showed an increase across all metrics that can be compared with June and July. The citation index saw the strongest growth: from 6% to 94% over 11 weeks.
| Credits | 24 June 2026 | 27 July 2026 | 10 September 2026 |
|---|---|---|---|
| Search queries mentioning the brand | 0 out of 18 | 17 out of 18 | 18 out of 18 |
| Citation index | 6% | 50% | 94% |
| Brand mentions | 0 | 10 | 19 |
| Models who are familiar with the brand | 3 out of 4 | 3 out of 4 | 4 out of 4 |
| Models who endorse the brand | 0 | 0 | 1 |
Source: three Aelo scans across four models: ChatGPT, Claude, Gemini and Perplexity. We are publishing this data with the client’s consent, without naming the company.
All three scans used the same frozen query set: 18 questions about building lime, agricultural lime, local deliveries and a regional manufacturer. We asked the questions without mentioning any brand names, that is to say, in a neutral scan. Each subsequent measurement of brand visibility in AI is comparable to the previous one only under these two conditions. The method of calculation is described in methodology, and the range of a single measurement is the password full scan.
Recognition means that the model is aware of a company’s existence and can describe it when someone asks for it by name. A recommendation means that the model itself suggests the company as a good choice when a customer asks about a category, e.g. „which manufacturer of fertiliser lime in the Łódź region”. These are two separate thresholds, and a brand can remain at the first one for a very long time.
The lime manufacturer featured in two surveys: in June and July, three models were familiar with it, but none recommended it. In September, all four were familiar with it, and one moved up to the second tier.
We describe this situation in more detail in the text He knows it, but doesn’t recommend it, and a situation where the model does not recognise the company at all, on the website ChatGPT doesn’t know my company. In this article, we break down the mechanism by which models select brands to respond to How AI chatbots choose the brands they recommend.
Models began to treat the manufacturer’s website as a source of information on lime, rather than simply as one of many sites in the search results. In June and July, for these search queries, they mainly cited the domains of major manufacturers and general-interest websites. In September, the brand’s website tops the list of sources: models cite it for 17 out of 18 search terms. The next domain, a classifieds website, appears for 11 search terms.
The report dated 10 September assesses the website as ready for citation. The brand has an extensive set of pages on building and fertiliser lime, a page on standards, a page about the company and a blog. Following the initial audit, not a single new subpage was created: the changes concerned whether the existing content was suitable for citation. The nature of these changes is described the first part of the case study.
The difference between a mention a quoting is key here. A mention is the brand name appearing within the text of a reply. A citation is a reference to the brand’s website as a source. In this case, citations grew faster than any other metric, and the website provided content that influencers could reference: specifications, standards, price and delivery range. We refer to the absence of such specific details as content gap.
In the September scan, the brand has 19 mentions. That is more than any other manufacturer of a comparable size: the nearest competitors have 15, 11 and 10 mentions. Only two international groups appear more frequently than it does, and their models are regarded as the benchmark for the entire lime category.
In July, the brand had 10 mentions and was trailing behind several domestic manufacturers. Six weeks on, it is outperforming all other players of its size. We explain in the article why models are more likely to mention companies that feature in multiple sources Why does AI recommend the competition?.
Perplexity mentions the brand in all queries, whilst Claude has been doing so since July; the brand’s average position on his lists has improved from 2.3 to 1.8. In practice, Claude usually lists the manufacturer as the first or second company.
| Model | 24 June | 27 July | 10 September |
|---|---|---|---|
| Claude | 0% | 100%, items 2 and 3 | 100%, item 1.8 |
| Perplexity | 72% | 67%, item 16.5 | 100%, item 11.2 |
Perplexity was the weakest link in July: the brand appeared in two-thirds of the responses and usually at the bottom of a long list. In September, it features in every response, and its average position has moved up by more than five places. This is important because Perplexity searches the web for every response and displays the sources, so it reacts quickly to changes in what can be found about the brand.
The results of the models vary from one measurement to the next. The text explains the reason for this variability Why does AI give a different answer every time?.
The brand performs best for search queries relating to building lime and lime mortar, particularly those associated with the Łódź Voivodeship and the region where the plant is based. These are the search terms for which the website provides the most specific details: compliance with the PN-EN 459-1 standard, product specifications and the plant’s location.
Search terms relating to agricultural lime are performing less well, even though the brand has separate local pages for them. That is why this area is next on the implementation plan.
The next stage mainly involves activities outside your own website, as these pave the way for recommendations in the other three models:
For the same search queries, the models draw on external websites: a job board, a classifieds site and agricultural websites. This is a list of places where the brand’s presence can lead to further recommendations. Tasks from the report are posted on the board in the module Tasks.
The effects of GEO are first evident in citations, then in recommendations. The citation score rose from 6% to 94% before the first recommendation appeared. The page, which the models can easily cite, works for every subsequent query. We explain what GEO is and how it differs from SEO on the page SEO, GEO and AEO: the differences and in the text Visibility in AI and SEO.
A recommendation follows a diagnosis; it does not replace it. In two surveys, the brand was recognised but not recommended. The first recommendation only appeared once all four models were familiar with it.
A trend only becomes apparent when a series of measurements is taken. Three screenshots showing the same 18 queries reveal a trend that isn’t apparent from a single screenshot. We explain on our website why we carry out measurements once a month How often should visibility be measured?, and why the effects take time to appear, on the website Why do changes take time?.
Between June and September, the models themselves also changed, so part of the growth cannot be attributed solely to work on the brand. We are describing a single case within a narrow category, not a promise of results. We carried out all three measurements using the same set of queries, without the brand name in the queries, across the same four models. We publish more of our measurements in the section Research, and we’ve summarised the basics of the topic in the guide Brand visibility in AI: how it works and how to measure it.
You can carry out the manual test yourself by following the instructions How to check whether AI recommends your brand. The module carries out regular measurements across four models using a fixed set of queries Visibility in AI. The Free account gives you 250 AeloCoins to get you started, plus a full brand report: mentions, citations, sources, brand recognition and barriers to be removed.
Agencies carrying out this type of measurement for clients will find further details on the following pages Aelo for agencies i Aelo for SEO agencies, and local businesses on the website Aelo for small businesses. The query limits for the plans are in price list.
A brand recommendation in AI refers to a situation in which a language model – such as ChatGPT, Claude, Gemini or Perplexity – spontaneously identifies a company as a good choice in response to a question about a category, without the company’s name being mentioned in the question. This is a higher threshold than a ‘mention’, where the model merely lists the brand, and higher than ‘recognition’, where it is able to describe the brand when asked about it directly. In the case of the lime manufacturer, the first recommendation appeared in the third measurement, on 10 September 2026.
Approximately 11 weeks: from the scan on 24 June to the scan on 10 September 2026. During this period, the brand went from zero mentions to appearing in all 18 search queries, and its citation score rose from 6% to 94%. The recommendation appeared last among the metrics, only after all four models were familiar with the brand.
No. In September 2026, all four models surveyed are familiar with the brand, and one recommends it. The other three mention it in their responses, but do not identify it as their recommended choice. Moving on to making a recommendation in subsequent models usually requires independent signals from outside one’s own website, as we describe in the text He knows it, but doesn’t recommend it.
This is because citations depend mainly on whether the website contains facts that the model can reference: product specifications, compliance with standards, price and delivery range. These elements are controlled by the company itself, on its own website. Recommendations, on the other hand, also depend on what others are saying about the brand, so they grow more slowly. In September, the manufacturer’s website was the most frequently cited source for queries relating to the brand: in 17 out of 18 search phrases.
Yes. All three scans used the same set of 18 queries, entered without brand names, across the same four models. Changing the set between measurements would have made the results incomparable, which is why the set was fixed from the first scan onwards. The caveat concerns the models themselves: they varied between measurements, so some of the increase may be due to these changes.
There is no guarantee that this pace will be maintained. The lime manufacturer operated in a niche sector; it already had a comprehensive website with several dozen local subpages and a recognisable product name. In a highly competitive sector, or where the website is less robust, results take longer to materialise. We describe the factors that influence the pace of growth on the page Why do changes take time?.
First, make your own website easy to reference: specific details about products, prices, coverage and standards, structured data, and content accessible to search engine bots. Then build your presence outside your own domain: customer reviews, listings in independent directories, and features on industry websites that search engines use when processing queries in your category. A list of such websites can be found in the sources cited by search engines, and we describe the process in the text Why does AI recommend the competition?.
Ask a few AI chatbots the questions that customers ask before making a purchase, without mentioning the company name, and repeat each question several times. Make a note of whether the brand appears, where it appears, alongside which competitors, and which sources the model uses. You’ll find full instructions, including a measurement sheet, in the text How to check whether AI recommends your brand, and a shorter version on the website How to check whether AI mentions the brand.
