Research

Original data on how AI assistants recommend small products, free to quote and reuse under CC BY 4.0, plus the outside research we rely on.

By Dasha. Updated November 1 2026. https://heardofyou.io/research

This is where Heard of You publishes original data on how AI assistants treat small products. Which ones get named, which pages get cited, and how much the answers change. Everything here is free to reuse under CC BY 4.0 with a link.

Why bother? It is the evidence behind what we sell. Nobody with an indie-founder audience publishes it. And original numbers are what other people quote, which is exactly what we help clients earn.

The first study is in progress and publishes the week of November 16 2026.

What we publish, and why

Every study on this page follows the same rules:
- A fixed, published prompt list, so anyone can repeat the runs.
- Every prompt run three times per assistant, because single answers do not repeat (SparkToro: the same brand list less than 1 time in 100).
- Mention rates, never ranks.
- Raw data as a downloadable CSV, with the assistant, model version, date, prompt, run number, products named and sources cited.
- Data collection dates and model versions shown on the page, and a limits section that says what the study cannot show.
- CC BY 4.0 on all of it.

A study costs hours, not money. One assistant query costs us roughly five cents in API fees. A 600-answer study is about $27 of compute and several days of reading answers.

Flagship study: do AI assistants recommend small products?

Status: in progress. Publishing the week of November 16 2026.

We asked 4 AI assistants (ChatGPT, Claude, Gemini and Perplexity) 600 buyer questions across 10 indie software categories. That is 50 real buyer questions, 5 per category, each run 3 times on each assistant.

The question is the one every solo founder asks first: does ChatGPT recommend indie or small products at all, or only the big brands?

The 10 categories. In each one, solo founders compete with incumbents:

1. Invoicing and finance for freelancers
2. Project management for small teams
3. Note-taking and personal knowledge
4. Email marketing for creators
5. Scheduling and booking
6. Form builders
7. Screen recording and screenshots
8. AI writing assistants
9. Habit and focus trackers
10. Website builders for indie makers

What the study will report, per category and per assistant:
- How often a small, independent maker's product was named at all, against the share of answers that named only incumbents.
- Recommended versus merely cited.
- Which kinds of pages were cited: "best X" lists, Reddit, review sites such as G2 and Capterra, YouTube, vendor sites, directories. That includes the answer to "do directories help?", which gets its own page in month two.
- How much the product list changed between the three runs.

What it cannot show: why a product was named, whether being named led to a sale, or anything about categories outside these ten. The study page prints those limits.

The prompts come from the prompt library (/prompts) template. The exact 50 are published with the data. The dataset will be a CSV under CC BY 4.0, with a short method page and a notebook on GitHub.

First Muse data

On November 1 2026 we added Muse, Meta's AI assistant, to our daily checks. Day one, across three small sites: one was named in 3 of 5 answers, two in none. What it searched, what it cited and what Meta says about its crawlers: What Meta's Muse recommends: first data from 3 small sites (/research/what-metas-muse-recommends).

Our own visibility log

Every month we run our own test prompts against ourselves and publish the results, good or bad: Heard of You visibility log (/research/heard-of-you-visibility-log).

Our first client case study starts at day 0, published before the work: a home dumpling business, named in 5 of 20 answers on day 0 (/research/daily-dumplings-case-study).

It starts near zero, because the site did not exist until October 31 2026. That is the point. A service that sells AI visibility should show its own.

Coming later

| Study | Question it answers | When |
|---|---|---|
| How often AI recommendations change | Are ChatGPT's product recommendations consistent in small categories? The same prompts, ten runs per assistant, three days apart | Month 2 |
| Does llms.txt matter? | Which bots actually request /llms.txt and .md pages, from our own server logs and those of pilot clients who agree | Month 2, after 60 days of logs |
| Do directories help AI visibility? | Pulled from the flagship data: how often directories appear among cited sources next to lists and reviews | Section of the flagship at launch, own page in month 2 |
| Citation maps by category | Which exact pages and domains each assistant cites for "best X" in a category, refreshed monthly | Three categories with the flagship, then one a month |

Data licence

All original data and charts on this site are published under Creative Commons Attribution 4.0 (https://creativecommons.org/licenses/by/4.0/). Reuse, adapt and republish, including commercially. Credit "Heard of You (heardofyou.io)" with a link to the study page. Third-party figures we quote belong to their authors and link to the original.

Third-party research we rely on

We did not discover how AI recommends products. These people did. Our guides and methods lean on the studies below most.

Vendor studies are marked. They come from companies that sell AI visibility, SEO or PR tools, so they have a commercial interest in the result. They tend to be large and correlational.

Peer-reviewed and academic
- Tannenbaum, 2026 (pre-print) (https://arxiv.org/abs/2609.23162): brands absent from the assistant's live search were named 2.8% of the time; with their site cited, 91.4%.
- Chen et al., University of Toronto, 2025 (https://arxiv.org/html/2509.08919): 72.7% of AI sources for US software questions were earned media.
- Chu and Hou, 2026 (pre-print) (https://arxiv.org/abs/2606.17443): identical specs, the known brand wins 100%; a rating edge under 0.1 stars flips it.
- Aggarwal et al., KDD 2024 (https://arxiv.org/html/2311.09735v2): citations, quotations and statistics raised visibility 30 to 40%; keyword stuffing did not.
- C-SEO Bench, NeurIPS 2025 (https://ar5iv.labs.arxiv.org/html/2506.11097): most content tricks were ineffective; being first in the model's context mattered most.
- Columbia, What Is Your AI Agent Buying? (https://ar5iv.labs.arxiv.org/html/2508.02630): ratings, review counts, badges and position drive agent picks.
- Wharton GAIL, 2026 (https://gail.wharton.upenn.edu/research-and-insights/technical-report-agentic-shopping/): a single Wirecutter or Reddit source can override a model's pick.

Industry and vendor studies
- SparkToro (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/): the same prompt repeats the same brand list less than 1 time in 100.
- AirOps and Kevin Indig (https://www.getpassionfruit.com/research/is-query-fan-out-a-marketing-buzzword-or-something-relevant): 58% of position-1 pages cited, 14% at position 10; 44.2% of citations from the first third of a page.
- Muck Rack, October 2026 (https://www.globenewswire.com/news-release/2026/10/28/3390134/0/en/generative-pulse-earned-media-still-drives-more-than-84-of-ai-citations.html) (vendor): over 84% of 207 million citations were earned; 0.3% paid.
- Ahrefs, 75,000 brands (https://ahrefs.com/blog/ai-brand-visibility-correlations/) (vendor): YouTube mentions correlate most with AI visibility.
- Ahrefs, best lists (https://ahrefs.com/blog/best-lists-research/) (vendor): listicles were 43.8% of cited page types.
- Peec (https://docs.peec.ai/research/most-cited-domains-by-engine) (vendor): most-cited domains per engine; Reddit and YouTube in every top five.
- GetIntel (https://getintel.ai/blog/best-saas-directories-2026/) (vendor): directories rarely cited across 9,978 answers.
- SE Ranking (https://seranking.com/blog/llms-txt/) and Ahrefs (https://ahrefs.com/blog/llmstxt-study/) (vendor): llms.txt shows no correlation with citations; 97% of files got zero requests.

Who uses which assistant
- G2, The Answer Economy (https://company.g2.com/news/g2-research-the-answer-economy): 51% of software buyers start in an AI chatbot more often than Google.
- McKinsey (https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/who-wins-when-ai-does-the-shopping-new-evidence-on-agentic-commerce): 53% of about 21,000 consumers have used AI for shopping.
- Supabase, State of Startups 2026 (https://supabase.com/state-of-startups): Claude is the top paid assistant among startup builders (59%).
- Stella Rising via Search Engine Land (https://searchengineland.com/real-people-actually-prompt-ai-geo-479731): how real people phrase prompts (n=524).

What to do next

Want to know where your own product stands before the study lands? The free check (/check) applies the same method to you: 5 buyer questions, 3 runs, 5 assistants, results in 2 working days.

Writers and researchers who want early access to the flagship data: email hello@heardofyou.io.

Questions

Can I use your data in my own article?

Yes. Everything here is CC BY 4.0. Quote it, chart it, build on it, and credit Heard of You with a link to the study page.

Why publish research instead of just selling the service?

The assistants read third-party pages. A tiny company becomes worth citing by producing numbers nobody else has. Also, founders deserve evidence, not a pitch.

Is the flagship study out yet?

Not yet. The runs are in progress and the study is scheduled for the week of November 16 2026. The method and categories are below. The numbers go up when they exist, not before.


Sources
- Tannenbaum, A fitted stage model of brand mention in AI search, 2026 (pre-print) (https://arxiv.org/abs/2609.23162)
- Chen et al., University of Toronto, earned media in AI search, 2025 (https://arxiv.org/html/2509.08919)
- Chu and Hou, brand bias and rating edges, 2026 (pre-print) (https://arxiv.org/abs/2606.17443)
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (https://arxiv.org/html/2311.09735v2)
- C-SEO Bench, NeurIPS 2025 (https://ar5iv.labs.arxiv.org/html/2506.11097)
- Columbia, What Is Your AI Agent Buying? (https://ar5iv.labs.arxiv.org/html/2508.02630)
- Wharton GAIL, agentic shopping technical report, 2026 (https://gail.wharton.upenn.edu/research-and-insights/technical-report-agentic-shopping/)
- SparkToro, AIs are highly inconsistent when recommending brands (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)
- AirOps and Kevin Indig, 815,484 query-page pairs (https://www.getpassionfruit.com/research/is-query-fan-out-a-marketing-buzzword-or-something-relevant)
- Muck Rack, Generative Pulse, 207 million citations, October 2026 (vendor) (https://www.globenewswire.com/news-release/2026/10/28/3390134/0/en/generative-pulse-earned-media-still-drives-more-than-84-of-ai-citations.html)
- Ahrefs, AI brand visibility correlations, 75,000 brands (vendor) (https://ahrefs.com/blog/ai-brand-visibility-correlations/)
- Ahrefs, best lists research, 26,283 source URLs (vendor) (https://ahrefs.com/blog/best-lists-research/)
- Peec, most-cited domains by engine (vendor) (https://docs.peec.ai/research/most-cited-domains-by-engine)
- GetIntel, 9,978 AI answers on directories (vendor) (https://getintel.ai/blog/best-saas-directories-2026/)
- SE Ranking, llms.txt across about 300,000 domains (vendor) (https://seranking.com/blog/llms-txt/)
- Ahrefs, llms.txt study (vendor) (https://ahrefs.com/blog/llmstxt-study/)
- G2, The Answer Economy (n=1,076) (https://company.g2.com/news/g2-research-the-answer-economy)
- McKinsey, Who wins when AI does the shopping (https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/who-wins-when-ai-does-the-shopping-new-evidence-on-agentic-commerce)
- Supabase, State of Startups 2026 (n=2,000) (https://supabase.com/state-of-startups)
- Stella Rising via Search Engine Land, how real people prompt AI (n=524) (https://searchengineland.com/real-people-actually-prompt-ai-geo-479731)