9 Buyers Found Us Through AI. Our Analytics Saw Almost None of Them.

• 9 Buyers Found Us Through AI. Our Analytics Saw Almost None of Them.
• 9 Buyers Found Us Through AI. Our Analytics Saw Almost None of Them.
• Published: October 24, 2026

Contributors: 9 Buyers Found Us Through AI. Our Analytics Saw Almost None of Them.
Published: October 24, 2026

Summary

Between May and October 2026, 9 buyers told upGrowth on recorded calls that ChatGPT, Claude, Perplexity or an AI agent put us on their shortlist, while GA4 logged AI-sourced form submissions on only 4 days. Repeat checks in ChatGPT and Perplexity on 22, 23 and 24 October 2026 found upGrowth named in 2 of 24, both times by ChatGPT on the same query, with the first mention sourced from 2 third-party roundups rather than our own site.

Between May and October 2026, 9 people who contacted upGrowth said an AI tool put us on their shortlist: ChatGPT, Claude, Perplexity or an AI agent they built themselves. Over roughly the same period, Google Analytics 4 recorded 1,375 sessions from AI assistants and only 4 days with a tracked contact form submission from those sessions. The AI search leads were real. The attribution was close to zero.

On 10 September we published a score of 0 out of 100 for our own AI visibility and promised to publish the next number, good or bad. This is that update. It covers what buyers told us on calls, what GA4 actually recorded, what the engines say about us today, and what we’d do differently if we were starting from scratch.

What happened since we scored 0 out of 100?

On 9 September 2026 our own AI Answer Gap tool ran “best B2B SaaS growth marketing agencies in India” through 8 engines. Every engine missed us. Score: 0.

On 22 October 2026 at 08:35 IST, we ran the same question in ChatGPT, logged out, with no account history. ChatGPT named upGrowth 9th in a list of 10.

ChatGPT’s description of us was “Best for: Performance-led growth and experimentation”, followed by a line we’d rather not have seen and are publishing anyway: “Confirm the depth of its experience in your specific SaaS category.” That is the exact weakness we flagged in September. Our case studies don’t show SaaS depth in a form an engine can extract, and the engine said so.

The more useful finding is where the mention came from. ChatGPT didn’t cite upgrowth.in for it. It cited 2 third-party roundups: Koda’s “Top 10 B2B lead generation agencies for SaaS and Tech Enterprises in India” (dated 11 August 2026) and Digital Patron’s “Top 15 B2B Marketing & Lead Gen Agencies in India (2026)” (dated 14 October 2026).

We didn’t pitch either publication and we didn’t pay for either listing. Both included us on their own.

That’s humbling. In September we said fix number 3 was getting into the third-party lists the engines read. Our own pages, which we control and work hard on, didn’t get us named. 2 lists we had no hand in did. The lesson isn’t “buy listings”. It’s that engines trust a second opinion more than your own, so the work is to be the agency other people write about.

Perplexity is a different story. We ran the same 4 buyer questions through Perplexity on the same morning, including the September baseline. upGrowth appeared in none of them.

We then ran the exact prompts our buyers told us they used. ChatGPT, logged out, missed us on all 3: “Suggest a performance marketing partner in India”, “Give me a list of the top 20 SEO agencies in India”, and “Which agency in India can help with both SEO and GEO?”.

That last miss stings. GEO is the core of our practice, and on 22 October ChatGPT answered a GEO question with 4 other agencies. 2 of them have “SEO plus GEO” or “AI SEO” in their brand or domain name.

Then we ran all 8 checks again on 23 and 24 October 2026, same prompts, same method.

On 23 October: 0 out of 8. ChatGPT’s answer to the baseline question had changed almost completely overnight. Of the 8 agencies it listed, only 3 had appeared in its 22 October list, and upGrowth wasn’t one of them.

On 24 October: 1 out of 8. ChatGPT put upGrowth back on the baseline list, this time 6th in its “My top picks”, and told buyers to evaluate us for “B2B SaaS references, sales-qualified lead quality, and CAC reporting.”

So the honest scorecard across 3 days is 2 mentions in 24 checks. Both came from ChatGPT on our hardest query, which named us on 2 of 3 days. The other 7 prompt and engine combinations named us 0 times in 21 tries, including every prompt our buyers told us they typed. That’s consistent with the October traffic drop below, and it’s the opposite of a victory lap.

A screenshot of an AI answer is a weather report, not a ranking. If anyone, including us, shows you a single screenshot as proof of AI visibility, ask for the repeat runs.

A note on method: we ran ChatGPT logged out. Perplexity, Gemini and Google AI Mode were signed in to my own account, and Perplexity personalised its answers even in incognito mode, so we’ve treated its results as indicative only and left Gemini and Google AI Mode out of the scorecard. Every check also came from a Pune IP address, and 2 answers added a “local Pune option” because of it.

Also Read: Your Competitors Are Writing the Lists That AI Reads

How did our AI search leads actually find us?

9 buyers told us on recorded calls between May and October 2026 that an AI tool surfaced upGrowth. These are their words from the call transcripts, lightly trimmed and not reworded.

Sarvagya Verma, The Stack (20 May 2026). He asked Claude to “give me a list of top 20 SEO agencies in the country and you guys showed up in that Excel sheet.”

Kaan, Smart Capital (17 June 2026). “I was from AI search, ChatGPT.” He was looking for help with both SEO and GEO.

Soniya Agrawal, Unifund (16 September 2026). “I searched in Claude for, uh, you know, performance marketing partner and got your result there.” She took the company name from Claude and then found me on LinkedIn.

Pierre Sernet, QWANZ (17 September 2026). “It was through my AI agent… We selected about 15 firms… narrowed it down to about 4. And you were one of them.”

Sandeep Saran, digiRM (22 September 2026). He used Perplexity: “I put my requirement there and… asked them to… find… right agencies. So yours was one of those agencies.”

Farooq Chisty, Brevo (5 October 2026). “I just asked Claude to just set up… an automation… it scouts for a bunch of agencies, scores them based on their… proof of work in Shopify and TikTok ads.”

3 more gave shorter answers when I asked how they found us. Dr Vijay Rai of LivoTale said “ChatGPT” (29 September). Dorji of Drukyul Holidays answered “Engine” when I asked whether it was Google or an AI engine (19 October). Sourabh of Prozpr said “maybe it was Claude because we use much” (12 October).

The first buyer to read your website is increasingly not a person. 2 of these 9 used AI agents that shortlisted and scored agencies before a human looked at any of them. Pierre’s agent cut 15 firms to 4. Farooq’s automation scored agencies on proof of work. If your site doesn’t state your proof in a form a model can read and compare, you lose at a stage you never see.

Why didn’t Google Analytics see our AI search leads?

GA4 recorded 1,375 sessions from AI assistants on upgrowth.in between 1 July and 21 October 2026, but only 4 days with a tracked contact form submission from those sessions. Most AI-sourced buyers don’t click through from the chat to a form. They copy the name, search for it, open LinkedIn, or book a call directly.

Here’s what GA4 shows for the window:

1. ChatGPT sent 573 sessions, Gemini 347, Claude 238, Perplexity 190, Microsoft Copilot 19 and Meta AI 8.
2. Organic search sent 46,640 sessions in the same window. AI assistants were under 3% of that.
3. AI visitors were more engaged. 26% of AI assistant sessions triggered a key event (360 of 1,365 in GA4’s AI Assistant channel), against 16% for organic search (7,591 of 46,640).
4. Most of those key events were clicks on “Contact Us”, not submissions. GA4 logged 292 contact clicks and 11 contact form submissions from AI sources.
5. Those 11 submissions fell on 4 dates. 8 of them came from a single Gemini session on 12 July.

Our contact form compounds the problem. It has no “How did you hear about us?” field. Every one of the 9 buyers above was identified only because I asked on a call and the call was recorded. Without the recording, all 9 would sit in our CRM as “website enquiry” or “direct”.

We also found a trap in the other direction. When we asked our AI meeting notes tool how a client found us, it returned a confident quote saying the client found us through ChatGPT. The transcript didn’t contain that sentence. The client had actually found us through a partner directory. If you’re measuring AI discovery from AI-generated summaries, you’re measuring the summariser’s guess.

AI discovery is real, and standard analytics undercounts it badly. Treat GA4’s AI channel as a floor, not a total.

Also Read: How to Measure AI Visibility: The Scoring Method We Use

Is AI traffic to upgrowth.in growing or falling?

Falling, sharply, in October. GA4’s AI Assistant channel sent 452 sessions in July, 455 in August, 337 in September and 123 in the first 21 days of October, which puts October on pace for about 180.

ChatGPT accounts for most of the drop. It sent 190 sessions in September and 20 between 1 and 21 October. Organic search held at roughly the same daily rate over the same period, so this isn’t a site-wide tracking failure.

We don’t know the cause yet and won’t guess in print. It could be a change in how ChatGPT shows links, a change in which of our pages it retrieves, or a shift in which competitor lists it trusts. What we can say is that the ChatGPT mention we saw on 22 October came from third-party roundups, not from our own pages, and our own pages are what sent the July to September traffic.

Which pages does AI send people to?

AI engines sent visitors to our data-heavy reference pages, not our service pages. The top AI landing pages between July and October 2026 were YouTube CPM guides, pricing benchmarks for influencer marketing, Instagram ads, LinkedIn ads and social media management, and our GEO and AEO pricing benchmark.

That pattern matters more than the totals. Engines cite pages that contain specific, current numbers a model can lift into an answer. A page that says “we deliver results” gives a model nothing to quote. A page that gives a dated 2026 price range for a specific service in India does.

1 AI landing page converted far better than the rest. Our social media management pricing page drew 26 AI sessions and 27 key events in the window, more key events than sessions.

What would we do if we were starting from zero today?

Start by instrumenting, not publishing. Add a source question to every enquiry form and ask the question on every first call. Without that, you’ll be told AI doesn’t work by a dashboard that can’t see it.

Then work in this order:

1. Add the form field. “How did you first hear about us?” with an option for “AI assistant (ChatGPT, Claude, Perplexity, Gemini)”. This costs 10 minutes and fixes the measurement problem above.
2. Ask on every call and record the exact words. The prompt a buyer typed is the most valuable keyword data you’ll ever get. “Performance marketing partner” and “top 20 SEO agencies in the country” came straight from 2 buyers.
3. Get into the lists the engines already read. Run your category question, open the sources the engine cites, and work out how to be on those pages honestly. In our 22 October test, 2 roundups did more for us than our whole site.
4. Publish numbers, not adjectives. Our AI traffic went to pages full of 2026 pricing and benchmarks. Proof of work with figures is what both human buyers and agents like Farooq’s scoring automation compare.
5. Fix the gap the engine names. ChatGPT told buyers to “confirm the depth” of our SaaS experience. That’s a content brief, written by the engine, for free.
6. Test logged out, on more than 1 engine, on more than 1 day. ChatGPT named us on 22 October, dropped us on 23 October and named us again on 24 October. In September, the same Perplexity query flipped results within 56 minutes. One check is a snapshot, not a score.

Also Read: We Scored 0/100 on Our Own AI Visibility Audit (Part 1)

What we’re not claiming

We’re not claiming AI search leads are our biggest lead source. By GA4 it isn’t, and by call notes it’s 9 conversations in roughly 5 months. We’re not claiming we can guarantee a ChatGPT or Perplexity mention for anyone, because we can’t guarantee one for ourselves, as Perplexity showed by missing us in all 12 of its checks. And we’re not claiming to know why October traffic fell.

What we are saying is narrower and more useful. Buyers are using AI tools, and increasingly AI agents, to build agency shortlists. Those buyers are mostly invisible to standard analytics. And the content that gets you onto their lists is specific, numerical, third-party-confirmed proof.

Common Questions About AI Search Leads

Your Next Move: Find Out Who the Engines Name Instead of You

Run your own buyer questions through the AI Answer Gap tool. You’ll see which engines name you, who they name instead, and which sources they trust. It’s free and needs a business email.

If you want the full picture, including the citation sources each competitor wins on and a plan to get onto them, book a 30-minute call. And yes, I’ll ask you how you found us.

Book a 30-minute strategy call.

About the Author: I’m Amol Ghemud, Chief Growth Officer at upGrowth Digital. We help SaaS, fintech, and D2C companies shift from traditional SEO to Generative Engine Optimization. That shift has driven a 5.7x lead volume increase for Lendingkart and a 7x increase in AI Overview ranking power for Vance.

About Author