There is no page two in an AI answer

I've spent a decade in growth. I ran it hands-on at ByteDance, Foundit, Mahindra Finance and IDFC FIRST Bank, and helped move more than $600M in revenue. Before that I co-founded an edtech company, Scholarnex, straight out of engineering school.

I say this not as a résumé but as a confession: everything I knew how to do was built for a world that is quietly ending.

The field nobody reads

Every growth team has a field on its sign-up form that asks, "How did you hear about us?" Most teams never read it. We did, because it was the only honest instrument we had left.

Our organic traffic had gone flat. Nothing in our dashboards had broken. Rankings looked fine. And yet one word kept showing up in that field, more often every month: ChatGPT.

These were real sign-ups, from a channel we couldn't see. We had no idea what ChatGPT was saying about us. We didn't know which questions it recommended us for, or who it named when it didn't name us. Every tool in our stack was built for ten blue links, and none of them could answer the only question that mattered.

Five names, then silence

Here is the shift, as plainly as I can put it.

Search gave you a page of ten results, and then a second page, and a third. If you lost, you lost by degrees. You could see where you stood and climb.

An answer engine doesn't work like that. A buyer asks for the best tool in your category, and the AI names three, four, five companies, and stops. There is no page two. There is no click to measure if you weren't named. You aren't ranked lower. You simply don't exist in that conversation.

That's a different kind of loss. You can't optimise what you can't see, and for most companies today, the shortlist is invisible.

What decides the shortlist

When we started measuring, the answers surprised even us.

AI rarely trusts what you say about yourself. In our research, 0.8% of AI citations pointed to the brand's own website. In one scan of category questions across four engines, 43.5% of the cited sources were comparison pages: listicles, alternatives and versus pages. Across just 80 answers, 387 different websites were cited. Your homepage is one voice among hundreds, and rarely the loudest.

And when we asked the engines about a brand-new company by name, only one engine in four described it correctly. The other three answered from what they already believed.

So the shortlist isn't decided on your website. It's decided by the reviews, roundups and comparisons the engines lean on, and by what they learned before you ever shipped your latest feature.

Honest numbers, or none at all

There is one principle I care about more than any feature, and it comes from years of sitting in growth reviews.

AI answers are noisy. Up to 27% of the names in an answer can change from one day to the next. Ask the same question twice and you might get a different list. If a tool shows you one answer and calls it your position, it's showing you a coin toss.

So we never trust a single response. We ask your buyers' questions many times, across ChatGPT, Claude, Perplexity and Google AI Overviews, and aggregate what comes back. Every number carries a confidence range. If a change is just noise, we'll tell you it's noise, and we won't take credit for it.

I've watched too many teams celebrate a lucky week. I didn't want to build another dashboard that helps people fool themselves.

Seeing isn't enough

A score on its own changes nothing. So Proofsource runs the whole loop:

• Visibility. Your buyers' real questions, asked every day, and who gets named.
• Why you lost. Every lost question, who won it, and what the answer leaned on.
• Citations. The pages AI actually trusts in your category, and which of them leave you out.
• Fixes and agents. Drafts built from the pages already winning. You approve, agents publish, verify the answers changed, and learn from every change.

Building it as operators

I'm building Proofsource with Amit, who ran growth at Uber and slice. We built the tool we needed and couldn't find. In private beta, five teams paid us before we opened the doors, and that told us we weren't the only ones staring at that sign-up field.

What I believe is simple. The next decade of growth won't be won by whoever ranks highest. It will be won by whoever the machines trust enough to name. That trust is earned the old way, by being genuinely good, and being written about honestly by people who aren't you. It's just measured in a new place.

If you want to know where you stand, check your AI visibility for free. There's no card needed. And if you tell me your category, I'll tell you who ChatGPT recommends for it today.