Do AI Engines Ignore Small Software?

In one audit of about 37,000 AI runs, roughly half of the lowest-profile brands the authors studied, startups and regional companies, never appeared even once. If an engine doesn’t name your small product, that is common at this size and not a verdict on your product. It is one preprint with limits, set out below.

What the audit did

Four researchers at Unusual.ai, a company that sells AI brand marketing, posted the preprint to arXiv in May 2026. As far as its arXiv page shows, it has not been peer reviewed. They ran 215 commercial prompts across 19 sectors on four model settings: OpenAI’s gpt-5.4-mini and Anthropic’s claude-sonnet-4-6, each at low and high reasoning effort, called through their APIs, not the ChatGPT or Claude apps. The models searched the web through their providers’ own search tools and, in some runs, through two outside search services (Exa and Brave). The authors then checked which of 533 brands showed up.

The brands came from outside lists and were sorted into five tiers. Tiers 1 and 2 are category leaders from G2 and Gartner lists. Tier 3 is mid-stage companies from Crunchbase. Tier 4 is startups from Y Combinator batches between winter 2023 and summer 2025. Tier 5 is regional companies from national business registers. The authors describe the tiers as a rough proxy for how well known a brand is in its sector, and say they do not claim it captures revenue, customer numbers or market share. Where the lists were thin, part of the catalogue was drafted with an AI model and reviewed by hand.

What they found

• The authors single out one setting: Claude Sonnet 4.6, high reasoning, its own web search. In it, a tier 1 brand surfaced in 77% of the queries in its own sector, tier 2 in 60%, tier 3 in 23%, tier 4 in 8% and tier 5 in 3%. “Surfaced” means it turned up in the search results the model retrieved, which is broader than being named in the final answer.
• Across all four settings and three search set-ups, 48% of tier 4 brands and 52% of tier 5 brands never surfaced in any of the roughly 37,000 runs in their sector.
• Even tier 1 brands, which surfaced in about three-quarters of queries in that setting, won only 25% to 41% of the recommendation slots they reached across the four settings. The authors conclude that for them the gap is differentiation, not visibility.

What this can’t tell you

The smallest tier here is funded startups, which are bigger than most solo projects, so a one-person tool probably sits at or below the bottom of this ladder. That is our inference, not something the paper measured. The audit ran on a single day, so it says nothing about change over time. It didn’t test Gemini, Perplexity or Google AI Overviews, and its prompts covered the US, UK and EU.

What the authors suggest

For tier 4 brands they say the data implies three channels. These are getting into authority lists, publishing canonical content on platforms their search services reach (Medium, Substack and GitHub README pages for developer tools), and being included in third-party comparison articles. That is their reading of the results, not a tested remedy, and their company sells this kind of work.

Check where you stand

Because answers vary from run to run, one try tells you little. The how-to-check guide covers doing it by hand, and why AI engines name your competitor instead of you covers the other evidence on who gets named. If your product is new, see how long until AI engines find your new website. The free check runs one real question across ChatGPT, Perplexity and Google AI Overviews with no account, and the live demo shows a real account over time.