The short version
- The market is large and still being divided: Menlo Ventures (opens in a new tab) estimates companies spent $37 billion on generative AI in 2025, $19 billion of it on applications, and that 76% of enterprise AI use cases are now bought rather than built.
- AI tools spread from the bottom up: Menlo found 27% of AI application spend comes through product-led growth, where individual users adopt first, against 7% in traditional software.
- Buyers are skeptical of AI claims: Gartner (opens in a new tab) estimates only about 130 of the thousands of vendors selling “agentic AI” are real, and calls the rest “agent washing.”
- New AI products are hard for assistants to surface: in a study of 112 Product Hunt launches (opens in a new tab), ChatGPT named them in only 3.32% of open questions such as “What are the best AI tools launched this year?”, though it recognized them 99.4% of the time by name.
- Recent pages carry weight in a fast-moving category: across each assistant’s dated citations in our freshness study, 17.4% to 22.6% were pages published in the last 90 days, compared with 6.9% of Google’s top 10.
Who pays for AI software, from a single seat to an enterprise contract?
Individual users often adopt first, then teams and enterprises buy, and the best-known products reach billions in revenue.
The category covers three kinds of company: AI-native applications (coding tools, writing and meeting assistants, support agents), generative AI model and platform providers, and established software companies adding AI features. Their buyers range from a single developer on a credit card to an enterprise committee. Firms that build custom systems for clients face a related task, covered in our guide for machine learning companies.
Menlo Ventures’ 2025 enterprise report, based on a survey of about 500 US enterprise decision-makers and a market model, gives the clearest picture of spend:
| Measure | 2025 figure |
|---|---|
| Enterprise spend on generative AI | $37 billion, up from $11.5 billion in 2024 |
| Spend on AI applications | $19 billion |
| Share of AI use cases purchased rather than built | 76% |
| AI deals that reach production, vs. traditional software | 47% vs. 25% |
| Application spend via product-led growth, vs. traditional software | 27% vs. 7% |
Menlo counts at least 10 products earning over $1 billion in annual recurring revenue and 50 earning over $100 million. It reports that Cursor reached $200 million in revenue before hiring a single enterprise sales rep, and that startups took 63% of the AI application market in 2025.
Consumer AI is a second market. Sensor Tower (opens in a new tab) reports AI app downloads grew 148% in 2025, and Menlo’s 2026 consumer report finds 55% of US AI users now pay for at least one AI product.
What a customer is worth depends on whether they stay. In RevenueCat’s 2026 report (opens in a new tab) on subscription apps, AI-powered apps earned 41% more revenue per payer but saw subscribers churn 30% faster. For AI companies, acquisition and retention are two separate problems.
Where does AI search sit in how buyers find AI tools?
At discovery and comparison, and for AI tools it often replaces the Google search a buyer would once have run.
Using AI to research a purchase is close to universal among business buyers: in Forrester’s (opens in a new tab) survey of nearly 18,000 of them, 94% reported using AI during their buying process. In G2’s 2025 survey (opens in a new tab), generative AI chatbots were the top influence on vendor shortlists at 17.1%, ahead of software review sites at 15.1%.
AI products are a large share of what those buyers purchase. In TrustRadius’s 2026 survey (opens in a new tab) of technology buyers, AI tools or tools with AI features made up 59% of the purchases buyers made in the past year, and 75% of buyers who purchased an AI tool said it lived up to their expectations. G2 found half of enterprise buyers at companies with 1,000 to 5,000 employees had switched vendors for better AI.
AI search also changes what “ranking” means. In our rankings study, only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question. A strong Google position for “best AI note-taker” does not guarantee a mention when the same question goes to an assistant.
Which questions do buyers ask when comparing AI tools?
Questions about use cases, alternatives, head-to-head comparisons, data safety, accuracy and price. We wrote the example prompts below to mirror how buyers compare AI tools; they are not drawn from real logs.
| Buyer concern | Illustrative prompt |
|---|---|
| Use case | “What’s the best AI tool for turning sales calls into CRM notes in HubSpot?” |
| Category | “Which AI coding assistants work best for a large TypeScript monorepo?” |
| Alternatives | “Alternatives to Jasper for a B2B content team that needs brand voice controls” |
| Comparison | “Cursor vs GitHub Copilot for a 40-person engineering team” |
| Data and security | “Does this AI search tool respect SharePoint permissions, and is customer data used for training?” |
| Accuracy | “Which AI support agents have published resolution rates, and how were they measured?” |
| Price | “How much does ElevenLabs cost for a podcast network, and what counts as usage?” |
These are questions where the answer changes month to month. Google documents that its AI features may use “query fan-out” (opens in a new tab), running several related searches before answering, and OpenAI documents that ChatGPT search (opens in a new tab) rewrites a question into targeted queries for its search providers. Our inference: for AI tools, those searches pull in recent launch posts, reviews, benchmarks and community threads, and old pages lose out to newer ones.
How does AI visibility turn into revenue for an AI company?
Through two paths: individual signups that spread inside companies, and enterprise pilots that convert unusually well.
The bottom-up path. A user asks an assistant for a tool, signs up, and brings it to their team. With 27% of AI application spend arriving this way, according to Menlo, an assistant’s answer to one person can seed an enterprise account. Menlo names n8n, ElevenLabs, Gamma and Wispr Flow among companies that scaled in this way.
The enterprise path. A buying group asks assistants to map a category, builds a shortlist and runs a pilot. Menlo found 47% of AI deals reach production against 25% for traditional software, so getting into the pilot matters a lot. These groups are larger than usual: Forrester found that purchases including generative AI features involve buying groups twice the size, 14 members against seven.
In both paths the assistant’s answer comes before the money, and the product has to keep the customer afterwards. We infer that for AI companies with high early churn, being recommended for the right use case matters more than being recommended often: a user who arrives for a job the product does well is more likely to stay.
Are AI companies competing with the assistants that recommend them?
Often, yes; the largest assistant makers sell AI products in many of the same categories.
Menlo’s data shows the overlap. Horizontal AI, the largest application category at $8.4 billion, is 86% copilots, “led by ChatGPT Enterprise, Claude for Work, and Microsoft Copilot.” Menlo’s 2026 consumer report adds that once most people use AI for a task, a product’s competition “becomes any AI that can do the job—including the general AI assistant the consumer already has open.”
So an AI writing, search or meeting tool may need ChatGPT, Gemini, Claude or Copilot to recommend it over the assistant’s own built-in features. Whether assistants favor their makers’ products has not, to our knowledge, been measured in a public study. We have no evidence either way and do not assume it.
What we can say is practical. A specialist product will be named when its advantage is specific and documented, such as a named integration, a measured accuracy figure or a compliance status, rather than when it claims to be generally better than a general assistant. That is our inference, not a platform rule.
Why do assistants recommend some AI products and pass over others?
Mostly independent, recent evidence about the product; the platforms do not publish how they choose.
What has been observed:
- Recognition is not recommendation. In Sharma’s Product Hunt study, new products were recognized by name almost every time but surfaced rarely in open questions. Perplexity found launches more often when they had more referring domains and Reddit presence. It is a single-author study run on a small ChatGPT model through a developer interface, so treat it as a signal rather than a verdict.
- Independent coverage counts. When more independent sites name a brand in the pages assistants cite, recommendations follow: in our brand-entity study, each tenfold increase went with 4.7 times the odds of being recommended.
- Newer pages win more citations. For the same questions, the pages the four assistants cited in our freshness study were about half as old as Google’s top 10, counting from first publication. It is a pattern, not proof that updating a page causes citations.
- Self-promotion is common and visible. In our study of “best X” lists, 24.2% of numbered lists that AI surfaces cited ranked their own publisher first. In AI software, where nearly every vendor publishes “best AI tools” lists, that kind of evidence is easy to discount, we infer.
Credibility is the trust factor that matters most here. Gartner’s “agent washing” warning and the US Federal Trade Commission’s Operation AI Comply (opens in a new tab) both target overstated AI claims. In that sweep, DoNotPay agreed to pay $193,000 over claims that it offered “the world’s first robot lawyer.” Buyers have reason to look for proof, and the assistants answering them search for reviews and named sources. For more on reputation, see how brands build authority for AI search.
What does GEO involve for a company that sells AI products?
Generative engine optimization (GEO) gives assistants accurate, current, independently confirmed evidence of what your product does and for whom.
For a company whose product is itself AI, the work usually looks like this:
- A precise identity. Say exactly which jobs the product does, for which users, on which models and integrations, with the same words on your site, docs, marketplaces and review profiles. Vague “AI platform” language is easy to confuse with a dozen rivals.
- Proof, not adjectives. Publish how accuracy, speed or savings figures were measured, link to customer stories with names, and keep security, data-use and training-data policies public and plain.
- Freshness. Date and update the pages that describe features, pricing and model support, so assistants do not repeat last year’s product.
- Independent coverage. Earn reviews on software platforms, coverage in trade press, comparisons by independent writers and real discussion in developer and practitioner communities.
- Honest comparisons. Explain where you differ from general assistants and named rivals, including where they are the better fit. Whether such pages earn citations is covered in do comparison pages help B2B brands get cited by AI.
- Presence across assistants. Check ChatGPT, Gemini, Claude, Copilot, Perplexity and Google’s AI features, because each is also a potential competitor and none behaves the same way.
No one can promise that an assistant will recommend a given AI product over its rivals, or over itself. GEO raises the odds that the evidence it finds is strong, current and accurate.
What can’t the research tell AI software companies yet?
Buyers clearly use AI to find AI tools; whether assistants treat their makers’ own products differently is still unknown.
- No public test of self-preference. We found no published study of whether ChatGPT, Gemini, Claude or Copilot favor their makers’ products in recommendations.
- Market figures are estimates. Menlo’s spend figures combine a survey with a market model, and Menlo invests in several companies it names.
- Vendor and platform interests. G2 and TrustRadius run review platforms; RevenueCat sells subscription software. Their data are useful but not neutral.
- Revenue links are unproven. No public study ties AI visibility to revenue for AI software companies. The broader question of whether AI visibility pays is reviewed in does AI visibility drive business results.
- Fast change. Assistants, models and the products they recommend change within months, so any snapshot ages quickly.
How should an AI software company test whether assistants are feeding its signups and pilots?
Ask the assistants your buyers use the same use-case questions your best customers asked before they bought.
That first check shows whether you are named for the jobs you do best, which rivals and general assistants appear instead, and whether your features, pricing and data policies are described correctly. Then fix the facts, publish the proof and earn the independent coverage that confirms it.
If your growth depends on signups that turn into teams, or on pilots that turn into contracts, talk to us about an audit of how assistants present your product. We will show how AI assistants describe your product against competitors and built-in assistant features, and which evidence would most improve your chance of being named for the use cases that bring in revenue. For the full scope, our generative engine optimization service page walks through how the diagnosis leads to proof pages, fresher product facts and independent coverage.
Frequently asked questions
Do AI assistants recommend AI startups or only big names?
They can recommend startups, but new products are rarely surfaced unprompted. When 112 Product Hunt launches were tested, ChatGPT recognized almost all of them by name yet put them forward in only 3.32% of open discovery questions.
Does ChatGPT favor OpenAI products over competitors?
No public study has measured this, so nobody can say. What is documented is that the major assistant makers sell AI products that compete with many AI software companies.
Why does my AI product show outdated features in AI answers?
Assistants draw on whatever pages they find, and older pages about your product may still rank. Dated, updated feature and pricing pages, plus current third-party coverage, give them newer evidence to use.
Is “AI-powered” in our messaging enough to be recommended?
No. With “agent washing” common, according to Gartner, and the FTC acting against deceptive AI claims, specific and checkable proof of what the product does carries more weight than the label.
Sources
- Menlo Ventures (2025-12-09), 2025: The State of Generative AI in the Enterprise (opens in a new tab)
- Menlo Ventures (2026-09-15), 2026: The State of Consumer AI
- Gartner (2025-06-25), Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (opens in a new tab)
- Federal Trade Commission (2024-09-25), FTC Announces Crackdown on Deceptive AI Claims and Schemes (opens in a new tab)
- Sensor Tower (2026-01-21), Boosted by Gen AI services, consumers spent more money in apps than games for first time (opens in a new tab)
- RevenueCat (2026), State of Subscription Apps 2026 (opens in a new tab)
- Forrester (2026-01), The State Of Business Buying, 2026 (opens in a new tab)
- G2 (2025), Proving Value in the Age of AI: 2025 Buyer Behavior Report (opens in a new tab)
- Demand Gen Report (2026-07-30), TrustRadius: AI Has Changed How Buyers Research, But Not What They Trust (opens in a new tab)
- Google Search Central (2025), AI features and your website (opens in a new tab)
- OpenAI Help Center (2025), ChatGPT search (opens in a new tab)
- Sharma (2025), The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries (opens in a new tab), arXiv:2601.00912.
- Underneath (2026), How fresh are the pages AI engines cite?
- Underneath (2026), Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?
- Underneath (2026), Do Wikipedia and schema make AI assistants recommend a brand?
- Underneath (2026), How many “best of” lists cited by AI rank their own brand first?