Do AI Assistants Recommend Well-Known Brands? We Checked 79 of Them

Established SEO and content agencies, asked about in their own category, were named in 43 of 171 AI answers. That is one answer in four. Of 19 agencies, 17 were named in fewer than half of the answers about what they sell, and 5 were never named at all.

Those are agencies people have heard of, in the one industry whose job is being found. The other three groups we measured did better, and the reasons they did better matter more than the numbers.

How the study worked

The sample

On 17 September 2026 we picked 79 companies across four groups:

• DTC and e-commerce brands (20)
• B2B SaaS companies (20)
• Marketing and SEO agencies (19)
• Mid-market brands with an in-house SEO team (20)

Every company was chosen by a person who already knew its name. That is a deliberate choice with a known cost, covered below.

The questions

For each company, a language model wrote three questions a buyer in that category might ask. Each question went to three AI engines: Perplexity, Gemini and DeepSeek. That gives nine answers per company.

A typical agency question looked like this: "best SaaS SEO agency for scaling organic traffic". We read every company's questions by hand before counting it. Seven companies got questions from the wrong category, so they were swapped for reserves.

What "named" means

An answer counts if it names the company. A passing mention counts the same as a top recommendation. This is the most generous reading available, and it still produced the agency result.

The results

A few answers failed to come back, which is why some totals are just under 180.

Agencies: the one clear result

Three of the 19 agencies sell AI search optimisation themselves. Leave them out and 5 of the remaining 16 were still never named.

The engines disagreed less than you might expect. Gemini named an agency in 17 of 57 answers, Perplexity in 14, DeepSeek in 12. No single engine explains the gap.

DTC: usable, with a caveat

7 of 20 well-known DTC brands were named in fewer than half of the answers about their own products. With the bias below in mind, that number is a floor, not an estimate.

SaaS and in-house SEO: not a finding

3 of 20 is a statement about these 20 companies. It does not say much about a smaller company in the same space, so we are not going to stretch it into one.

Why these numbers are kinder than your own

The sample was established names

A brand that comes to mind easily is more likely to be named by an AI engine too. So every "fewer than half" above is a lower bound. A less known company in the same category starts further back.

The questions were written by a model that knew the brand

This is the bigger bias, and we found it by reading the questions. A model that recognises a brand tends to describe that brand's own niche. A bedding company gets asked about the exact fabric it sells, not about "sheets". A narrow question has fewer possible answers, and the brand is one of them.

Agencies may have been helped less by this, since "best SEO agency" stays a crowded question however narrowly it is phrased. That could be part of why their result is so different. We can't separate the two effects with this data, and we won't pretend to.

Nine answers is a small window

AI answers vary between runs, even for the same question. Nine answers per company is enough to see a pattern across a group of 20. It is not enough to grade one company, which is why no company is named in this post.

How to check your own brand in 15 minutes

Step 1: Write three buyer questions

Write them the way a buyer who has never heard of you would, not the way your homepage describes you. Include one broad question ("best project management tool for agencies") and one narrow one.

Step 2: Ask three engines

Paste each question into Perplexity, Gemini and one more assistant. Use a logged-out or private window, so your own history doesn't steer the answer.

Step 3: Count the nine cells

Mark each answer as named or not named. For example, if you are named in 4 of 9, you are in the "fewer than half" group. Look at which questions the misses came from: broad or narrow tells you different things.

Then run the same nine checks again a week later. A single run tells you less than the change between two runs.

What we are doing with this

We run the same check on ourselves every week, across all five engines we track. In our latest run (14 September), our own brand was named in 0 of 90 answers about our category. That's why we built the product, and it's the number we are working to move.

If you'd rather not paste 9 questions by hand, the no-signup scan runs the same shape: three questions, Perplexity, Gemini and DeepSeek. A free account runs six questions across all five engines, ChatGPT and Claude included.

One difference, and it comes from this study. Reading the questions showed how narrow they were, so we changed how they are written: they now describe the category a buyer would recognise rather than a brand's own positioning. On these same 79 brands that moved the share of answers naming the brand from 62% to 53%, which means the counts in the table above would be lower if the study were re-run today. They are a floor twice over.

Method note: run on 17 September 2026. Raw answers are kept privately, because they describe other companies. We publish counts only, never a company name or a company's result.