# We asked five AIs about 41 indie products. 20 were never named.

2,940 AI answers about 41 small software products, October 31 to November 8 2026: how often each was named, by which AI, and who was named instead.

By Dasha. Updated November 13 2026. https://heardofyou.io/blog/we-asked-five-ais-about-41-indie-products

Between October 31 and November 8 2026 I ran my free check on 41 small software products. Each one got five buyer questions, asked three times, on five AIs. Across 2,940 answers, 20 of the 41 products were never named once. The median product was named in 1 answer out of 75.

This is not a new experiment. It is the data I already had from a week of free checks, added up. I made no new AI calls for this post. The products stay anonymous: no names, no domains, only totals. The big names that were recommended instead are fine to name, so I do.

**Video: 41 small software products, five AIs: the short version** (22 seconds, no sound, https://heardofyou.io/video/41-products-five-ais.mp4). A 22 second summary of 2,940 AI answers about 41 small software products: each dot is one product, 20 were never named, 3 in 4 were named in 2 answers or fewer, and the same AI gave the exact same list 4% of the time.

On screen: I asked five AIs about 41 small software products. The five assistants appear: ChatGPT, Gemini, Claude, Perplexity and Muse. Five questions a buyer would really ask, three times each, 2,940 answers. Each dot is one product, lit by how often it was named: 3 or more times, 1 or 2 times, or never. 20 were never named, not once. Half the room, invisible. 3 in 4 were named in 2 answers or fewer; the top 3 took more than half of all mentions. Same AI, same question, three runs: 4% of the time it gave the exact same list. One screenshot proves about as much as a coin that landed heads once. Which side of the 20 are you on? Heard of You.

## What I counted

Each check works the same way. I read the product's site and write five questions a buyer might type, one of each shape: best for a use case, under a price, what people like this actually use, alternatives to a big name, and a problem described in plain words. Each question goes to ChatGPT, Gemini, Claude, Perplexity and Meta's Muse, three times, with web search on where the API has it. That is 75 answers. A small model labels each answer (was the product named, was it recommended, which other products came up, which sites were cited), and I read every report before it goes out.

Nine of the 41 checks ran on October 31, before I added Muse. Those have four AIs and 60 answers. The other 32 have 75. I kept both and say which is which where it matters.

The sample: 41 small software products, mostly picked from Show HN and Indie Hackers launch posts in late October and early November 2026, plus a few founders who asked for a check. Left out: my own products, two well-known products I used as a calibration test, four local businesses (reported at the end), one non-software service, and checks that errored or had not finished. Where a product was checked twice I kept the standard check.

## Most were named almost never

| Named in | Products |
|---|---|
| 0 answers | 20 |
| 1 or 2 | 11 |
| 3 to 5 | 4 |
| 6 to 15 | 5 |
| 16 or more | 1 |

So 31 of the 41 were named in two answers or fewer. Pooled, the 41 products were named in 130 of 2,940 answers, which is 4.4%. When they were named, they were almost always recommended (124 of the 130).

The few that did get named took most of it. The top three products account for 67 of the 130 answers that named anyone in the sample. The best one was named in 39 of 75.

For scale: my own service was named in 0 of 343 answers on November 6. I am in the 20.

For contrast, a small dumpling shop in Florida that I track the same way was named in 93 of 179 answers on November 9 ([its log](/research/daily-dumplings-visibility-log)). Local questions, and a site with a page that answers each one.

## Which AI named small products most

| AI | Named | Out of | Rate | Products named at least once |
|---|---|---|---|---|
| Muse (Meta) | 38 | 480 | 7.9% | 11 of 32 |
| ChatGPT | 36 | 615 | 5.9% | 13 of 41 |
| Gemini | 26 | 615 | 4.2% | 5 of 41 |
| Perplexity | 17 | 615 | 2.8% | 5 of 41 |
| Claude | 13 | 615 | 2.1% | 4 of 41 |

Muse named small products most often, and ChatGPT named the widest spread of them. Two honest footnotes. Muse only ran on 32 of the checks, and 11 of its 38 hits were one product. On the 32 checks where all five ran, five products were named only by Muse and by no other AI. Claude and Perplexity were the hardest to get into, each naming four or five of the 41 at all.

## The question decides more than I expected

| Question shape | Named | Out of | Rate |
|---|---|---|---|
| Under a price ("best X under $20 a month") | 44 | 558 | 7.9% |
| Best for a specific use | 48 | 618 | 7.8% |
| What people like this use | 16 | 588 | 2.7% |
| Alternatives to a big name | 16 | 588 | 2.7% |
| A problem in plain words | 6 | 588 | 1.0% |

Specific questions named small products about three times as often as the persona and alternatives questions. The plain-words problem question, the one that sounds most like a real person, named them in 1 answer in 100. When the question is vague, the answer reaches for the names everyone knows.

## Who got named instead

Almost every answer named somebody. Only 72 of the 2,940 answers named no product at all. Most of the names were the obvious leader in each product's own category, so the long list is not very interesting. What is interesting is which names turned up across many different categories:

| Named instead | Checks it appeared in (of 41) |
|---|---|
| Notion | 13 |
| Claude | 10 |
| ChatGPT | 9 |
| Slack | 8 |
| Zapier | 7 |
| Obsidian | 7 |
| Google Sheets | 7 |
| Gmail | 6 |

Notion showed up for 13 different products. And in 10 of the checks, an answer suggested Claude itself as the tool to use, in 9 ChatGPT. A small product is not only competing with its direct rivals. It is competing with "just use Notion" and "just ask an AI".

## Your own site gets read when you are named

A product's own site was cited in 105 of the 2,940 answers, 3.6%. That sounds small, but look at it the other way round: of the 130 answers that named a product, 96 also cited that product's own site. The AIs read your homepage once they have decided you belong in the answer. Your homepage alone rarely gets you into the answer.

## Ask three times, get three different lists

For each product, each question and each AI, I had three runs. That is 980 sets of three.

- The list of products named was exactly the same in all three runs 41 times out of 980, about 4%.
- The first product named was the same in all three runs 396 times, 40%. That counts the name exactly as written, so "Notion" in one run and "Notion AI" in the next counts as a change.
- Where a product was named at least once in a set of three, it was named in only one or two of the three runs 44 times out of 67.

So if a founder asks ChatGPT once and does not see their name, that tells them less than it feels like. And if they do see it once, the same. This is why I count out of 75 and do not give anyone a rank.

## Local businesses were a different world

I also ran the same check on four local service businesses. They were named in between 19 and 48 of 75 answers. Four businesses in one region is not a study, only a contrast. Local questions get local answers, and the field of candidates is small. A Mac app is up against every Mac app.

## How I handled corrections

The labelling model makes mistakes, and the known one is the same-name trap: a product with a common name gets counted when the AI meant something else. In one check, a rival app had the same name; the model counted 6 hits and I cut it to 3 by hand. The numbers here use the corrected labels. For every check I confirmed that the per-answer file and the report's headline agree.

## Limits

- 41 products is a small number, from one week, picked by hand from launch posts and people who asked. Launch-week products are the newest and least written about, so read this as a floor, not a market rate.
- Five questions per product, written by me from each product's site. A different five would give different numbers.
- These are API answers with web search, from a US location, fresh each time. The apps people use may answer a little differently.
- Nine checks had four AIs, not five. Pooled rates mix the two; per-AI rates use each AI's own count.
- Splits by price or category would leave groups of a handful, so I did not make them.
- I sell a service that works on exactly this problem, and these free checks were also how I met founders. Weigh that as you like. The [table behind this post](/data/indie-products-ai-check-2026-11.csv) is there so you can check my sums.

If you want your own row, the [free check](/check) is the same five questions, three runs and five AIs, with who was named instead and which sites were cited. A person checks every report.

## Sources

- The anonymised table behind this post (CSV, one row per product) (/data/indie-products-ai-check-2026-11.csv)
- What 60 answers from four AIs look like when nobody has heard of you (our own baseline) (/blog/what-60-answers-look-like)
- Heard of You visibility log (/research/heard-of-you-visibility-log)