Plenty of people talk about "AI search" in theory. Two companies have shared real numbers in public: Tally, a bootstrapped form builder, and Vercel, the platform behind Next.js. Neither is a Substitutr customer. Everything below comes from their founders' own posts, linked in the sources at the end.
Tally: from a viral badge to AI recommendations
For its first five years, Tally grew through word of mouth and a simple loop: free forms carry a "Made with Tally" badge, someone fills out a form, sees the badge and signs up. Then the main source of new users changed, and Tally documented it in its regular milestone posts.
- June 2025, $2M to $3M ARR in four months. Co-founder Marie Martens posted that "AI search became our biggest acquisition channel" and that ChatGPT, Perplexity and others were driving the majority of new signups. The blog post says ChatGPT became their #1 referral source, with over 2,000 new users a week signing up via AI tools, "and those are just the ones we could track."
- April 2026, $5M ARR. The viral loop was "no longer our biggest acquisition channel. AI-powered search is." ChatGPT, Claude, Gemini and Google's AI Overviews were all growing as referral sources, with Claude growing fast.
- September 2026, $6M ARR and 2.5 million users. 43% of new users now discover Tally through ChatGPT, Claude, Gemini, Perplexity and other assistants.
Why it happened, in Tally's words
Tally's own theory: when ChatGPT turned on web browsing by default, "forums, Reddit posts, blog mentions, and authentic UGC" became part of what the AI reads. Tally already had years of users recommending it in exactly those places. In the April 2026 post they put it simply: word of mouth and AI search "are powered by the same thing: our community", happy users sharing the word and recommending Tally on Reddit.
How they measure it
Tally asks every new user where they found it, in a short onboarding survey. That matters, because analytics alone undercount AI: about the 2,000 weekly signups they could track from AI tools, Tally wrote that the survey shows "the actual number is a lot higher." A common reason: people read a recommendation in ChatGPT, then type the name into their browser, and show up as "direct" traffic.
Vercel: ChatGPT went from under 1% to 10% of signups
Vercel's CEO Guillermo Rauch shared the numbers on X as they happened:
- March 7, 2025: "ChatGPT now refers 4.8% of new @vercel signups and it's growing *very* fast. It was less than 1% 6 months ago."
- April 10, 2025: "ChatGPT now refers 10% of new @vercel signups, which have also accelerated."
In a June 2025 blog post, Vercel described how it now writes for AI search. In short:
- Find a frontier concept: watch X, Reddit, GitHub, Discord and forums for questions people are starting to ask, and look for gaps where competitors are shallow or absent.
- Publish the definitive, evidence-based source: original data, benchmarks and customer stories that are hard to copy.
- Structure for machines: consistent terminology, clean heading hierarchy, Schema.org/JSON-LD markup, and short, self-contained insights that are easy to cite.
- Seed authentic citations and set a refresh cadence so content stays current.
What the two stories have in common
These are our observations, not claims either company made:
- The product was already good. Both had years of real users before AI search took off. AI assistants amplified existing word of mouth; they didn't create it from nothing.
- Other people talked about them in public. Reddit threads, forums, blog posts, GitHub. That's the material assistants read when they answer "what's the best form builder?" or "where should I deploy my Next.js app?".
- They measured it early. Tally with an onboarding survey, Vercel by tracking signup referrers. Both noticed the shift months before most companies did.
What a small app can copy this week
- Ask "How did you hear about us?" at signup, with ChatGPT, Claude, Gemini and Perplexity as options. It's the cheapest way to see AI traffic your analytics label as "direct".
- Check your referrers for chatgpt.com, perplexity.ai, claude.ai and gemini.google.com.
- Help happy users talk about you in public. A "made with" badge, a community, honest answers in Reddit threads about your category. One real recommendation is worth more than ten posts on your own blog.
- Write pages that are easy to quote: a plain one-line description, a pricing page, comparisons and an FAQ, with clear headings.
- Find out where you stand today. Ask the assistants what they recommend in your category, or run a free AI visibility check to see whether they name you and who they name instead.
A caveat: all numbers here are self-reported by the companies, and they come from two products with strong communities. Results for a brand-new app will be smaller and slower. But the direction is the same for everyone: when people ask an AI which app to use, you want to be in the answer.
For the data behind the shift, read how people find apps in 2026.
