Research / Edition 2026-09
50 tracked vendors were never named by any of ten AI models
Key findings
- 551 vendors were tracked across 33 categories. 501 were named at least once. 50 were never named in any of 1,650 answers from ten models.
- Absent from AI coding agents: Amp, Kiro, Zed and Factory. Absent from AI code review: Ellipsis, Sourcery, Bito and Baz.
- Absent from LLM observability: PromptLayer, Humanloop, Laminar and Sentry. Absent from product analytics: Plausible, Statsig and Umami.
- Absent from newsletters: Loops, Resend, Paved and Medium. Resend leads transactional email at 100% share and is never named as a newsletter tool.
- Moving from four models to ten cut the never-named list from 116 to 50. The 50 that remain are absent even with 50 answers per category.
50 of the 551 vendors tracked in the September 2026 Memetik Index were not named in any answer. Each category’s vendor list was built from the products a buyer would recognise, then extended with every product the ten models actually named. The 50 are real products with real customers that ten current models from four providers, asked five buyer questions each, did not mention once in 50 answers.
Which vendors were never named?
| Category | Never named |
|---|---|
| AI coding agents | Amp, Kiro, Zed, Factory |
| AI code review | Ellipsis, Sourcery, Bito, Baz |
| LLM observability | PromptLayer, Humanloop, Laminar, Sentry |
| Newsletter platforms | Loops, Resend, Paved, Medium |
| Link-in-bio tools | Bio.site, Tap.bio, Snipfeed |
| Product analytics | Plausible, Statsig, Umami |
| Short-form clipping | Podcastle, Crayo, 2short |
| Browser automation | Apify, Notte |
| Creator monetization | Ko-fi, Buy Me a Coffee |
| AI music | Riffusion, Splash |
| Paid communities | Geneva, Tribe |
| Payroll | Paycor, Papaya Global |
| Transactional email | Customer.io, Courier |
| AI app builders | Rocket, Create |
| One each | Vapi (agent builders), Descript (avatars), Reve (image), Wispr (meeting notes), Freepik (video), Dorik (website builders), SambaNova (inference), Latenode (automation), Kill Bill (billing), Insightly (CRM), Convex (databases) |
Are these small products?
Some are. Many are not. Zed and Kiro are shipped by well-funded teams. Sentry is a large public-facing company; its LLM monitoring did not register in observability. Statsig was acquired by OpenAI; it did not appear in product analytics, including in OpenAI’s own three models. Resend leads transactional email with 100% answer share and 56% named first, and was never named as a newsletter tool. Vapi is one of the two best-known voice-agent platforms and did not appear under agent builders.
What the absent vendors have in common
A crowded incumbent set. In observability the models name LangSmith, Langfuse, Arize, Braintrust and Confident AI and stop. In product analytics, Amplitude, Mixpanel, PostHog, Heap and Pendo. Credible newer entrants get no slot.
Category mismatch. Resend is filed under transactional email, not newsletters. Descript is filed under voice and clipping, not avatars. Vapi is filed under voice agents, a category the prompts did not ask about. The models have a taxonomy, and a vendor sits in one cell of it.
Recency. Amp, Kiro, Factory, Notte, Reve and Wispr are products that grew in the past year. Ten models with web search on still did not surface them.
What changed with ten models
The first pass of this edition ran four models and left 116 vendors unnamed. Extending to ten models and 50 answers per category named 66 of those. The 50 that remain are absent at five times the sample, which makes each one a firmer finding than the four-model list was.
What this means for a vendor
A zero in this table means that when a buyer asks any current AI model the question your marketing is built to answer, your name is absent from the reply. The lever is presence in the sources the models cite for that category: documentation, engineering blogs and comparison pages for developer tools; YouTube and Reddit for consumer software.
Limitations
Vendor lists are ours; a product we did not track cannot appear as absent. Alias matching is exact-word, so a product referred to only by an unusual nickname could be under-counted. Fifty answers per category: a zero means fewer than one in fifty. One edition.