What 60 answers from four AIs look like when nobody has heard of you

Our own baseline, November 6 2026: the free-check slice, 5 questions x 3 runs x 4 AIs, published unedited, zeros included, with who was named instead.

By Dasha. Updated November 9 2026. https://heardofyou.io/blog/what-60-answers-look-like

On Friday November 6 2026 I ran the check I offer founders for free, in its four-assistant version, on my own six-day-old service. Five buyer questions, three runs each, on ChatGPT, Gemini, Claude and Perplexity: 60 answers. I was named in 0 of them. This page shows what that looks like in full, because a founder deciding whether to request a check deserves to see a real one, and because I said I would publish my numbers whether or not they flattered me. They did not flatter me. I had a forecast of zero and the forecast was not brave.

Day 0 for Heard of You is November 6 2026. This is the baseline. By then the site had a handful of plain-answer pages, some only two days old. Nothing has been done off the site yet. The day 30, 60 and 90 re-runs will be added as rows on the visibility log (/research/heard-of-you-visibility-log), not edited into this page. Sixty is the free-check slice; the full 405-answer run is further down.

The five questions and the 60 answers

These are the five starred prompts from the 27 I published before running anything. One per question shape, in the words a founder might actually type. They are locked. They will never be edited, because I know what I am like in January.

| # | Shape | Prompt |
|---|---|---|
| 1 | Best for use case | best way to get my SaaS recommended by ChatGPT |
| 4 | Price-capped | AI visibility audit under $200 |
| 7 | Persona | I'm a solo founder and ChatGPT never mentions my product, what should I do |
| 10 | Alternatives | alternatives to Peec AI for a one-person company |
| 20 | Problem first | why doesn't ChatGPT recommend my product |

These five are also the 20 cells (5 prompts x 4 AIs) my own promise is counted on. The target for day 60, January 4 2027, is the baseline plus three, floor three. So: 0 of 20 today, target 3 of 20.

| | ChatGPT | Gemini | Claude | Perplexity | Total |
|---|---|---|---|---|---|
| Answers | 15 | 15 | 15 | 15 | 60 |
| Named Heard of You | 0 | 0 | 0 | 0 | 0 |
| Recommended (not only cited) | 0 | 0 | 0 | 0 | 0 |
| Cited a heardofyou.io page | 0 | 0 | 0 | 0 | 0 |
| Searched the web before answering | 15 | 6 | 6 | 15 | 42 |

Models, as pinned on November 6 and recorded in every row: ChatGPT gpt-6-luna, Gemini gemini-3.8-flash, Claude claude-sonnet-5-5, Perplexity perplexity/sonar. Location: United States. Fresh request per run, no memory, no system prompt beyond the vendor default. The request parameters are in the CSV. Mention rate: 0 of 60, or 0%.

Who was named instead, and where the AIs read it

This is the table a founder should actually look at. When the AIs did not name me, they named someone. Counted across all 60 answers, once per answer:

| Named instead | Answers (of 60) | ChatGPT | Gemini | Claude | Perplexity |
|---|---|---|---|---|---|
| Otterly | 21 | 1 | 6 | 10 | 4 |
| Peec AI | 11 | 0 | 3 | 6 | 2 |
| Profound | 8 | 0 | 0 | 8 | 0 |
| G2 (a review site, not a product) | 8 | 0 | 8 | 0 | 0 |
| Capterra (a review site, not a product) | 8 | 0 | 8 | 0 | 0 |

Average number of products named per answer: 3.9. Answers that named no product at all and gave advice instead: 23. (Prompts 7 and 20 are questions, not shopping lists, so advice-only answers are expected there.)

The companies above are real and I have no relationship with any of them. They appear because a founder asking these questions would hear their names. There is no editorial here about whether they are good. They are on the list. I am not on the list.

Every answer that searched the web cited pages. Across the 60 answers there were 512 citations to 350 distinct pages on 188 domains.

| Domain | Citations | Kind of page |
|---|---|---|
| openai.com | 111 | OpenAI's own help pages on how ChatGPT search works |
| apify.com | 31 | directory |
| sona.com | 16 | other |
| google.com | 12 | other |
| dev.to | 9 | roundup |
| youtube.com | 8 | video |
| semrush.com | 7 | vendor |
| peec.ai | 7 | vendor |
| rankability.com | 6 | other |
| seofai.com | 6 | other |

By kind, the honest answer is that most of it does not fit the boxes. Directories were 6%, vendors' own sites 5%, "best tools" roundups 3%, review sites 2%, YouTube 2%, Reddit 1%, news under 1%. The other 82% were pages my classifier calls "other", and the biggest single chunk of that was ChatGPT citing OpenAI's own help pages 111 times, because three of the five questions are about ChatGPT itself.

heardofyou.io was cited 0 times in these 60. (It was cited in 43 answers across the whole run, almost all of them on the questions that name Heard of You. That is the assistants reading my homepage when asked about me, which is not the same thing as recommending me.)

This table is the to-do list. It is also, almost exactly, the deliverable I sell: get onto the pages in it, honestly, one at a time. The 25 I am pitching in November were chosen from this table and from the full 27-prompt run below, and the pitch log will be published with the day 30 row, rejections included. I expect rejections to be most of it.

One answer in full, and the branded check

The rule for which answer gets quoted was fixed in advance so I cannot pick a flattering one: the first answer in run order that names Heard of You, or if none does, the first answer to prompt 1 on ChatGPT, run 1.

> Prompt 1, ChatGPT, run 1, November 6 2026, model gpt-6-luna.
>
> The best route is to make your SaaS genuinely useful for a specific problem people ask ChatGPT about--and make sure ChatGPT can find and verify it. There’s no guaranteed way to get an organic recommendation; OpenAI says search placement isn’t guaranteed. (help.openai.com (https://help.openai.com/en/articles/9237897-chatgpt-search?utm_source=openai))
>
> ### Start with these steps
>
> 1. Allow ChatGPT search to crawl your site. Check that OAI-SearchBot isn’t blocked by robots.txt, your CDN, or bot protection. It’s separate from GPTBot, which is for potential model training--not search visibility. (developers.openai.com (https://developers.openai.com/api/docs/bots))
>
> 2. Make your site easy to understand. Clearly explain who the product is for, what it does, pricing, integrations, limitations, and how it compares with alternatives. Keep key pages public and readable without a login. This is practical discoverability advice, not a ranking formula.
>
> 3. Build independent credibility. Earn relevant customer reviews, case studies, and thoughtful coverage from trusted publications or industry sites. A product page can explain your claims; independent sources help substantiate them.
>
> 4. Consider an in-ChatGPT app only if it fits the product. If users could benefit from using your SaaS *within* a conversation--rather than just hearing about it--explore building and submitting an app. OpenAI describes chat-native apps as a way to complete useful workflows, but publication or recommendation isn’t assured. (openai.com (https://openai.com/index/developers-can-now-submit-apps-to-chatgpt/))
>
> 5. Test and measure. Ask ChatGPT common buyer questions in your category, with and without your brand name. Note whether your site or independent sources appear, fix gaps, and track referral traffic from ChatGPT.
>
> If you only do one thing this week: check crawler access, then publish one genuinely useful, detailed page for the exact problem your best customers are trying to solve.
>
> Sources it cited: help.openai.com (https://help.openai.com/en/articles/9237897-chatgpt-search?utm_source=openai), developers.openai.com (https://developers.openai.com/api/docs/bots), openai.com (https://openai.com/index/developers-can-now-submit-apps-to-chatgpt/).

(Unedited, except that the three long dashes in the original are shown as double hyphens, because this site does not print them.)

Prompts 23 to 25 ("Heard of You reviews", "what is heardofyou.io", "is Heard of You legit") and the ten branded prompts added for the runner are not visibility. An AI that names you when asked about you by name has told you nothing. They are an accuracy check. Of 88 branded answers:
- 34 did not find me. They asked which "You" I meant, or answered about a different Heard: a bookkeeping service for therapists, a hearing clinic, an app that reads coding-agent updates aloud. Claude (15 of its 22) and Perplexity (15 of 22) did this most. Claude: "no company, product, or site named Heard of You".
- 11 said something wrong. The wrong things were:
- The free check's size. Gemini (six times) and ChatGPT once said it is 4 assistants and 60 answers. It is 5 assistants and 75. The 60 is this very page's title, from before the fifth assistant was added, so the assistants were partly quoting me. Gemini also put the audit at 300 answers, the old four-assistant number. It is 375.
- Claude thought Heard of You is a music app. Three times, for "what is heardofyou.io", it described a project that links songs to personal memories, with Spotify support. That is a different thing with the same name.
- The guarantee. Gemini once called me "HeardOf AI" and said what I guarantee is the measurement. What I guarantee is the work: every listed deliverable ships by day 90 or the fee comes back, and if we miss the agreed target at day 60, I keep working for free until day 120.
- The other 43 described the service more or less as it is. ChatGPT got 19 of its 22 right.

Wrong facts get fixed at the source where I can, which is my own pages, and logged where I cannot.

The full run, and how stable it was

The 60 above are a slice of the baseline. The full run was all 27 published prompts, 3 runs, 4 AIs through their APIs with web search on, plus the 14 branded checks and the Spanish and Russian test prompts (404 answers). On the same evening I ran the 23 published questions that do not name me through Google search three times each, signed out, from the United States, and saved the AI Overview each time: 67 answers, because on two of the 69 searches Google showed no AI Overview at all. The table counts only the 23 English questions that do not name me.

| | ChatGPT | Gemini | Claude | Perplexity | Google AI | Total |
|---|---|---|---|---|---|---|
| Answers | 69 | 69 | 69 | 69 | 67 | 343 |
| Named (the 23 English questions that do not name me) | 0 | 0 | 0 | 0 | 0 | 0 of 343 |
| heardofyou.io cited | 0 | 0 | 0 | 0 | 0 | 0 |

The same row, with the per-engine split and the CSV, is on the visibility log (/research/heard-of-you-visibility-log). The log is the source of truth; this page is the story of the first row. A second log (/research/daily-dumplings-visibility-log), for a small dumpling shop tracked the same way, now sits next to it for contrast.

Three runs per prompt exist because one answer is not a measurement. In SparkToro's 2,961-prompt test the same prompt gave the same brand list less than one time in a hundred. Across my 113 prompt-and-engine pairs, the three runs agreed on whether I was named in all 113 (easy, when the answer is "no" three times). They agreed on the full list of products named in 18 of 113, and on the first product named in 36 of 113. That last number is why there are no ranks on this site. "You are number two" is a sentence about one Tuesday.

What surprised me, what happens next, and the limits

I expected zero on the non-branded prompts, and that is what happened, which is the least satisfying kind of right. I expected the branded prompts to either draw a blank or describe the homepage, and mostly they did. The thing I did not expect was that the most common wrong fact about me came from me: an older number on my own pages, repeated back by two assistants as if it were current.

What happens next is the work the sources table asks for: 25 pitches to the roundups and review sites it names, two follow-ups each, then stop; three honest pages on this site, including a comparison that says when to pick the competitor; questions answered on Reddit under my own name; no paid placements, no persona accounts, no bought reviews. The same five questions run every morning as a pulse, and the full 27 run every Monday. The day 30 row goes up on Monday December 7 2026.

Judge agreement, for the record: the small model that labels each answer flagged 39 answers for a second look, all of them on the branded questions and none in the 60 above. I read all 39, one at a time, against my own pages. The judge had counted 7 answers as knowing me when they had not found me, and it marked 5 answers wrong that were right, including the guarantee and the audit, because the fact sheet I gave it described the guarantee badly. The fact sheet is fixed. The branded numbers above are the reviewed ones. The human at the end of the pipeline is me. The human had coffee.

Limits: sixty answers is enough to tell "never" from "sometimes" from "usually" and not enough to measure a small change. The API answers may differ a little from what a signed-out person sees in the apps; the models are recorded and an app-versus-API comparison is scheduled for November. A US location was fixed throughout. Being named more often is the thing I can measure. It is not revenue, and I will not claim revenue I cannot see.

If you want to see what your own answers look like, the free check (/check) is the same five shapes and three runs, now on five AIs (Muse joined after this baseline), with who was named instead and the sources cited, by email within 2 working days. A person checks every report. Or run it yourself with the do-it-yourself guide (/guides/check-if-chatgpt-recommends-you); it takes about an hour and costs nothing but the hour.

Sources
- Heard of You visibility log (the prompts, the cadence, the monthly rows, the CSV) (/research/heard-of-you-visibility-log)
- SparkToro, AIs are highly inconsistent when recommending brands, 2026 (consultancy with a tracking-vendor partner) (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)
- Tannenbaum, a fitted stage model of brand mentions in AI search, 2026 (pre-print, one study, GPT only) (https://arxiv.org/abs/2609.23162)