Before you hire a GEO (generative engine optimization) agency in India, ask for one thing: a measured before and after on the same buyer questions, weeks apart. GEO Close, a Mumbai AI search visibility studio, re-asks a fixed battery of real buyer questions each cycle. For the Singapore art brand Tangled Monk, direct AI recommendations went from 8 to 16 in one month (10 Jul to 24 Aug 2026), share of voice rose from 7.9% to 18.8%, and nothing regressed. That is the bar. Here is how to tell an agency that can prove movement from one that cannot.

1. They measure like-for-like, with your brand name never in the question

The only honest visibility test is whether the AI names you unprompted. A credible agency asks a fixed set of buyer questions (your brand name never in the question), 3 times per engine, with the same AI judge every cycle, so month-over-month is comparable. Ask to see the exact prompt list and two dated runs. A one-time screenshot with no baseline gives you nothing to measure against later.

2. They work all four levers, not just content

Being named by AI takes four levers, and a real agency names all four and tells you which it has done:

  • Technical and retrievability: product schema on every page, a complete sitemap, and AI bots (GPTBot, ClaudeBot, Google-Extended, PerplexityBot) unblocked. For Tangled Monk this covered all 59 pieces and 36 products missing from the sitemap.
  • Identity and entity: one consistent set of facts across Organization schema, Google Business Profile and Wikidata, so every source the AI checks agrees. For S. Nihar Exports, identity consistency moved from 4.0 to 14.0 out of 20.
  • Content built to be cited: answer-first pages, each held to a citability bar before publishing. One Tangled Monk page's citability rose from 1.3 to 7.3.
  • Off-site corroboration: independent listings and editorial mentions, so the AI is not relying on your own site alone.

3. They separate live-grounded answers from training memory

An AI can know your name from training data and still never cite you in a live search. The number that counts is the live one. For Tangled Monk, 15 of the 16 recommendations were live Gemini web citations, the retrieval behind Google AI Overviews, not training-memory recall. Ask which engine ran a live search, and what share of the wins were live.

4. They show the raw answers

Every claim should be checkable. GEO Close retains every raw AI answer from every run, so a share-of-voice figure resolves to the actual transcripts behind it. A score you cannot trace to real answers is a vanity metric.

5. They tell you what they have not done yet

The most useful line in Tangled Monk's month-one report was the honest one: the off-site lever, footprint 11 of 27 with 0 independent editorial pieces, had not been started, so the doubling came from on-site work alone. An agency that names the upside it has not touched is measuring reality. One that claims everything is done is selling a number.

The one question that settles it

Ask: show me a client you moved on the same prompts, weeks apart, with the raw answers. If they can, you are looking at measured GEO. If they cannot, you are buying hope. GEO Close has done exactly that across specialty chemicals (S. Nihar Exports, 0 to 21.4% share of voice), natural food colourants (THS IMPEX, 4.1% to 14.3%) and handmade art (Tangled Monk, 8 to 16 recommendations). See the full method on our approach page, or read the Tangled Monk case study in full.

Want the same read on your brand? Get a free GEO audit and see exactly who the AI names instead of you today.