9 Questions to Ask AnyoneDoing Your AI Search Work
When an agency claims they handle AI search or AEO, what does that actually mean?
Here are nine questions with concrete answers to test your agency, a prospective vendor, or us. The point isn't to catch anyone out. It's that "we're on it" and "here is our prompt set" are very different answers, and until now most CMOs have had no way to tell them apart.
Yohann JohnFounder, The Inner LabsPublished 28 Aug 2026 · Updated 6 Sept 2026
Key Takeaways
“We're on it” and “here is our prompt set” are different answers. Every question below has a concrete one.
The strongest test is not what a provider measures but whether they can show you the sources AI engines cite in your category.
If nobody can tell you your share of voice against your top three competitors, nobody is measuring.
A single monthly check is not measurement. AI answers vary between runs.
English-only reporting is incomplete for most Indian consumer categories.
A plan confined to your own website addresses a minority of what the model reads.
3.Which sources are the AI engines citing when they answer questions in my category?
Good answer
A list of specific domains, with frequencies — the citation share behind your category's answers.
Warning sign
A general answer about “authoritative sources” or “quality content.”
Why it matters: This is the actual diagnosis. Brands appearing instead of you are appearing because of specific sources. If your provider can't name them, they're guessing at the fix.
5.Are you measuring Hindi and Hinglish, or only English?
Good answer
Either “yes, here's the split” or “we tested it for your category and it made no difference, here's the comparison.”
Warning sign
They haven't considered it.
Why it matters: Indian consumers query across English, Hindi and transliterated forms. Different phrasings can surface different sources and return different brands. Measuring one and reporting it as your visibility is measuring part of the picture.
6.Can you break my results down by city and buyer type?
Good answer
Yes, and they show you.
Warning sign
One national number.
Why it matters: A single aggregate percentage can't tell you that you're strong in Mumbai and absent in Hyderabad, or that you appear for premium buyers and never for price-led ones. Those are the findings that change what you do next.
7.What are you doing off my website?
Good answer
Specific work on specific third-party sources. Named publications, aggregators, review platforms, communities.
Warning sign
The entire plan is on-site content and schema markup.
Why it matters: In the scans we run, third-party sources frequently account for a substantial share of the citations behind an answer about a brand. A plan confined to your own domain addresses a minority of what the model reads.
They know when it happened, and they can tell you what moved.
Warning sign
Blank look.
Why it matters: Engine updates shift citation patterns regardless of anything you do. Somebody tracking this can distinguish a platform change from a performance problem. Somebody who isn't will attribute both to whatever they happened to be working on.
9.Can I see the raw data?
Good answer
Yes.
Warning sign
Only a monthly slide.
Why it matters: A summary you can't audit is a summary you have to take on trust. If the underlying runs exist, there's no reason not to show them.
On the answers you'll get
Most agencies will answer three or four of these well. That's not incompetence. AI search measurement is genuinely new, and an agency doing solid SEO work has not necessarily built the infrastructure to run daily prompt sets across multiple engines, several languages and several cities.
The problem is the gap between "we've got it covered" and what's actually happening. Your agency may be doing sensible foundational work that helps. They may also be describing a checklist item rather than a programme.
You now have nine ways to find out which.
A note on us
We'd fail some of these too, depending on what you ask about. We don't hold SOC 2. We're a young company. If you're a US enterprise with an internal SEO team who just needs data, there are better-established options.
What we'd answer well is questions 3 through 7, which is where the Indian specifics live and where most providers built for other markets have not needed to invest.
It helps, and AI engines do draw on pages that rank well. But the work diverges: SEO optimises your pages to rank, AI visibility is largely about presence in the third-party sources a model reads. The skills overlap; the centre of gravity doesn't.
Should I move this work away from my agency?
Not necessarily. Some agencies are building real capability. The useful step is finding out where you actually stand, then deciding. The questions we’ve provided above are how you find out.
What if my agency gets defensive?
That's information. Every one of these has a straightforward answer if the work is happening.
How do I know if my agency is actually doing AI search work?
Ask questions 1, 2 and 3 above. Which prompts, what is my share of voice, and which sources are being cited. All three have specific answers if the work is happening. If the responses are general, the work probably is not.
Can an SEO agency learn to do this?
Some are building the capability. The infrastructure is the hard part: running daily prompt sets across multiple engines, several languages and several cities, and logging every cited source. That is a different operation from ranking reports, and it is worth asking what an agency has actually built rather than what they plan to.
How much should this cost?
It varies enormously by prompt coverage, languages, locations and whether execution is included. Ask what you're getting rather than comparing headline numbers. A cheap plan tracking 50 English prompts once a week and an expensive one tracking thousands of daily variants are not the same product.
Yohann John is the founder of The Inner Labs, an AI discovery and answer engine optimization platform for brands across India and the GCC. He works with retail, fintech, real estate and hospitality brands on how AI assistants describe and recommend them.
