How We Are Getting Yamm Labs Into ChatGPT, Gemini and Google AI Mode Answers: 90-Day Log, Day 1 Baseline

By Yamm Labs · Published 18 September 2026 · 9 min read

Why we are publishing this AI visibility log

We sell generative engine optimisation (GEO) to financial brands. The honest way to sell it is to run it on ourselves, in public, with the baseline numbers included, and to publish every monthly re-query whether or not it moves. This is the first entry. It is written on the day of the baseline, 18 September 2026, and it is deliberately unflattering, because a Day 1 that looked good would be a Day 1 nobody should trust.

Yamm Labs is a design-led brand agency for fintech and BFSI companies, founded in Gurugram in 2017. Clients include TATA AIA, HDFC Life, Axis Bank, Paytm Money and Aditya Birla Sun Life Mutual Fund. Yamm Labs is not related to Yet Another Mail Merge (YAMM), the Google Workspace add-on, or to yaM Labs, a US meeting-software startup. As you will see below, the engines had not worked that out.

How we tested on Day 1

• Date: 18 September 2026.
• Engines tested: ChatGPT, Google Gemini and Google AI Mode. Perplexity, Microsoft Copilot and Claude showed sign-in walls for logged-out use and were not tested; they join the matrix at the first monthly re-query.
• Conditions: logged out, fresh session per prompt, India IP, no personalisation.
• Prompts: discovery prompts across our three clusters (fintech and BFSI branding, GEO for BFSI, and outsourced PPAP documentation), plus two brand-fact prompts about Yamm Labs itself.
• Scoring: Present (named and described correctly), Partial (named with an error, or only in a citation), Absent (not there). For every run we recorded which brands were named and which pages were cited.

How to read the Day 1 tables

Each row in the tables below is one prompt run on one engine, not an average. Present, Partial and Absent describe what a buyer would have seen in that single answer. A Partial is not half a win. An answer that names Yamm Labs with the wrong city, or mixes us up with another company, can do more harm than an absence, because the buyer leaves with a wrong fact.

Besides the score, we record the brands each engine named and the pages it cited. The brands show who we compete with inside the answer. The cited pages show where the engine’s view of the category comes from, and those pages are the part we can influence: by publishing citable content, correcting listings or earning a place on independent lists.

Discovery prompts and brand-fact prompts measure different things. Discovery prompts ask for a recommendation without naming us, so they test whether we are on the engine’s shortlist at all. Brand-fact prompts name us directly, so they test whether the engine’s picture of Yamm Labs is correct. A brand can pass one and fail the other.

Why logged out, with a fresh session per prompt? A signed-in account carries history and preferences that can change which brands appear. Logging out and testing from an India IP is the closest we can get to a first-time buyer in our market. The trade-off is that some engines would not answer without sign-in, which is why the Day 1 matrix covers three engines rather than six.

Discovery prompts: who the engines named instead of us

Yamm Labs was Absent from 12 of 12 discovery runs. The table lists the runs, the brands each engine named and the sources it cited. Ratings and review counts are as displayed by the engine on the day.

Three things stand out. First, the same three or four agencies appear across all three engines for the fintech branding prompt, and they are the agencies that appear in the same two listicles. Gemini’s answer was, in effect, one competitor’s blog post read back to us. Second, for the Gurgaon prompt Google AI Mode did not read agency websites at all; it read Google Business Profiles, and ranked by rating and review count. Third, for the NFO prompt the engine went to large advertising and PR networks and framed the answer around regulatory compliance, which is exactly the framing our NFO work uses and which our site did not state in a liftable form.

Brand-fact prompts: what the engines think Yamm Labs is

Both brand-fact prompts scored Partial, with entity errors on every engine that answered.

The correct facts, for the record: Yamm Labs was founded in 2017 and is headquartered at DLF Garden City, Phase 1-A, Sector 92, Gurugram (Gurgaon), Haryana, India. The Ghaziabad address is the registered office as listed on a company registry site; the Mumbai line came from a LinkedIn experience entry that reads “Jul 2019 – Jul 2019, Mumbai”; and the yaM Labs confusion comes from a name collision with an unrelated company that has a Crunchbase and TechCrunch footprint. None of these errors were invented by the engines. Each one traces to a source we had never corrected, and to the fact that our own website did not state the founding year or the city in plain prose anywhere.

Technical gate: what was fine and what was broken

Already fine. robots.txt blocks only /wp-admin/. llms.txt is live. OAI-SearchBot, PerplexityBot and Claude-SearchBot all receive HTTP 200 from the site. This is why Technical GEO is our one Strong category on Day 1.

Found and fixed, or fix in progress, on Day 1:

• The Rank Math sitemap index was returning 404. Fixed in configuration; the fix is pending a rewrite flush, so we will confirm it on the next check.
• The homepage had three H1 elements and two conflicting Organization schema blocks. Reduced to one H1 and one Organization node.
• ZoomInfo lists our head office in Palo Alto. Correction requested.
• The founder’s LinkedIn experience entry for Yamm Labs reads “Jul 2019 – Jul 2019, Mumbai”. Being corrected.
• The Company Check gives the registered office as Ghaziabad. The registered address is a legal record; what we can do is make the operating address unambiguous everywhere else.
• There was no About page and no Contact page. Both were published today.

Day 1 scorecard

This is our own grading of our own site, using the six-category scorecard we use for clients. The weighted result is roughly 35 out of 100.

What we changed today

Everything below was published or fixed on 18 September 2026.

• Published an About page and a Contact page, each stating the founding year, Gurugram address and the disambiguation line about YAMM and yaM Labs.
• Reduced the homepage to one H1 and added an entity paragraph stating who Yamm Labs is, in prose an engine can lift.
• Replaced the two conflicting Organization schema blocks with a single ProfessionalService node carrying founding date, address, phone, email and sameAs links.
• Published a fintech and BFSI branding category page, a Gurgaon page, an NFO launch page and a cost guide, each answer-first with question headings.
• Published the GEO service page for BFSI brands, which describes the programme this log is measuring.
• Published five FAQ hubs: fintech branding (120 questions), BFSI compliance (80), GEO and AEO (60), PPAP documentation (80) and company questions (40), each with FAQPage schema.
• Added page excerpts to llms.txt so AI fetchers get a curated map of the site.
• Set noindex on two template pages that had no business being cited.
• Fixed the sitemap configuration (pending rewrite flush) and requested corrections at ZoomInfo and on LinkedIn.

That is the Days 1 to 30 pillar, technical and entity foundation, started on Day 1. Days 31 to 60 are citable content beyond what shipped today, and Days 61 to 90 are third-party authority: listicle inclusions, directory listings, reviews and source remediation.

What we cannot control

• The engines. They change sources, models and sampling without notice, and answers vary between sessions. A re-query that shows movement may be noise; a re-query that shows none may be lag. We report both as they come.
• Competitors’ content. The listicles the engines are citing belong to other agencies, who are under no obligation to include us. We will pursue independent lists and directories, not competitor pages.
• Third-party turnaround. ZoomInfo, The Company Check, LinkedIn and Crunchbase-type sources correct on their own schedules or not at all. The registered office on the registry is a legal record we cannot change by asking.
• Reviews. We will ask past clients for Google reviews. We will not incentivise them or buy them, so the count moves at the pace clients choose.
• Sign-in walls. Perplexity, Copilot and Claude could not be tested logged out on Day 1. If that persists, their baseline will be taken from a logged-in session and labelled as such.
• The sitemap fix. It is applied but not yet verified live; if the rewrite flush does not clear it, that is the first item on the next check.

90-day targets

These are the numbers we will be graded against on the Day 90 re-query. They are targets, not promises; no one can guarantee placement in an AI answer.

Common mistakes when measuring AI visibility

• Testing while signed in to your own account and reporting the result as what buyers see.
• Changing the wording of a prompt between checks, which makes any movement impossible to read.
• Counting a Partial as a win when the engine got a fact about the brand wrong.
• Fixing the site first and recording the baseline afterwards.
• Reading one re-query as a trend. Answers vary between sessions, so a single change may be noise.

How to start an AI visibility log for your own brand

The method in this AI visibility log works for any financial brand. These are the steps in the order we run them.

1. Write the prompts first. Pick the questions a buyer types before they know your name, grouped by service line, plus two or three brand-fact prompts that name you. Freeze the wording so every re-query is comparable.
2. Fix the conditions. Log out, open a fresh session for each prompt, and note the date and the country of the IP. Record which engines refused to answer without sign-in.
3. Record more than the score. For every run, write down Present, Partial or Absent, the brands named and the pages cited. The citations become the to-do list.
4. Trace every entity error to a source. A wrong city, a wrong founder or a name collision usually comes from a directory, a registry page or an old profile. List each source and what it says.
5. Check the technical gate. robots.txt, llms.txt, whether AI crawlers receive a 200 response, the sitemap, and the schema on the homepage.
6. Grade and record the baseline before changing anything. A baseline taken after the fixes is not a baseline.
7. Re-run the same prompts on a fixed date and publish the result whether it moved or not.

If your brand is regulated, the pages you publish to become citable still go through your normal review. Confirm the current text of any advertising rule with your compliance team before a page goes live.

Next check

Next check: first working day of October 2026. We will re-run every prompt on all six engines under the same conditions, publish the table in the same format, regrade the scorecard, and list what we are changing next. If the numbers have not moved, that entry will say so.

Related reading: the AI visibility (GEO/AEO) service page for BFSI brands, the GEO and AEO FAQ, our earlier best fintech branding agencies in India post, and the SEBI, IRDAI and RBI advertising rules guide that governs how we write pages for regulated brands.

Last updated: 18 September 2026