This Marketing Expert Kept Losing Client Traffic to ChatGPT. So He Built a Way to Measure It.

Himanshu spent years doing the thing every agency does. Rank the client, report the rankings, renew the retainer. Then the reports stopped making sense.

Impressions were holding. Positions were holding. Clicks were falling. Clients started asking why a page sitting at number one for a money keyword was sending less traffic than it did last year, and the standard answers didn't cover it.

The reason turned out to be simple. Buyers had stopped scrolling a page of options.

In They were asking ChatGPT, Perplexity, or Google's AI for a recommendation, getting three or four names back, and picking one. If the client wasn't in that answer, the client wasn't in the running, and nothing in any SEO dashboard recorded the loss.

Bisht says the fix he built around that problem lifted client traffic 185%. What's more useful than the number is the method, because it's replicable by anyone reading this without buying anything.

The Answerrank Method

We start with something boring for most people: write down the questions a buyer asks right before they choose someone in the category, then ask those questions to the assistants and log who gets named.

Not brand-name searches. Nobody who hasn't heard of you types your brand name. The questions that decide purchases look like "best invoicing tool for freelance designers," "X vs Z," "is X worth it," "cheapest way to do Y."

Run twenty of those across ChatGPT, Perplexity, and Google's AI mode and you get a competitive set that rarely matches the one in the client's deck. It usually includes two companies nobody at the client had heard of, and excludes the rival they've been obsessing over for two years.

Then comes the part that changed the strategy. Look at what those answers cite. It's almost never the brand's own landing page. It's Reddit threads, comparison posts, review roundups, YouTube reviews, Quora answers. The assistant is summarizing a conversation that's happening about you somewhere else.

There is a big factor of trust and reputation in this process. Because AI engines now also analyse if brand is trustworthy or not.

Why this is winnable

Classic SEO had a moat, and it was backlinks. You could write the best guide on the internet and lose to a worse one on a domain with 40,000 referring domains. Years and budget won that fight, which is why small operators mostly lost it.

Answer engines don't sort by domain rating. They assemble from what they can quote confidently, so a small brand that shows up in the right thread and publishes a fair comparison page can land in the same shortlist as a funded incumbent. That's an open field.

It's open because most incumbents aren't working on it yet. They're still buying keywords. Bisht's argument, laid out in his guide to tracking AI search rankings, is that manual spot checks give founders a false sense of security: answers shift week to week and change with the phrasing of the question, so a single check tells you almost nothing. Measure the same questions repeatedly or don't claim you're measuring.

The context makes the urgency clearer. Pew Research Center tracked the browsing of 900 US adults and found people clicked a normal result on 8% of searches showing an AI summary, versus 15% without one. Clicks on links inside the summary: 1%. A widely reported 2026 analysis put Google searches ending with no click at roughly 68%. Google has disputed the Pew methodology, and one study is one study, but anyone watching a Search Console graph flatten while impressions climb recognizes the shape of it.

What the work actually looks like

The fixes are less exciting than the diagnosis, which is probably why they're available.

Publish the comparison page that answers "us vs them" honestly, including where you lose, because a model will quote a fair comparison and skip a brochure. Answer the buyer question in the first sixty words of the page instead of after three paragraphs of positioning. Show up in the community threads the answers already cite, as a participant rather than a link dropper. Go after long-tail questions where no brand owns the answer yet, which is where a small team can win in weeks instead of years.

None of that requires software. Twenty questions, three engines, once a week, a spreadsheet, about thirty minutes.

Bisht eventually turned his version of that spreadsheet into AnswerRank, which runs buyer questions across ten engines, tracks how often the answer names you, shows which competitor is becoming the default, and returns a prioritized fix list. There's a free GPT SEO checker for a single-page read without signing up.

The honest takeaway isn't the tool. It's that the twenty-question test costs an afternoon, most people won't run it, and somewhere in your category a competitor is quietly becoming the default answer while everyone stares at rankings that no longer predict revenue.