How to vet an AI SEO agency: 12 questions before you sign

"AI seo agency" runs about $45 a click in the US and grew roughly 190% year over year, per Google Ads data pulled in September 2026. When a category gets expensive before it matures, opportunistic offers follow. Here is how to sort them inside a single call.

On measurement

• How many prompts do you measure me on, and how often?
• Which engines, and does that include Google AI Mode and AI Overviews?
• What do you do when an engine returns no generated answer for a query: do you count that as my brand being absent, or as a coverage gap?
• Do you measure my position inside the answer, or only whether I am mentioned?

The third question is the most revealing and almost nobody asks it. Counting a query with no generated answer as your brand being absent drives your score down mechanically without any competitor having taken your place. An agency with no opinion on that is not measuring seriously.

On action

• Who writes to my store, and into which fields exactly?
• What happens if someone on my team edits a product while you are working on it?
• Do you verify after writing that my customer-facing page did not change?
• Do you produce content, and if so with what factual validation?

Questions six and seven expose the engineering standard. Catalog work without write-conflict control eventually overwrites someone's work, and you will find out late.

On proof

• How do you demonstrate that a fix produced an effect?
• Do you record a baseline before intervening?
• What do you do when the observed gap is too small to credit to your work?
• Can you tie revenue to this visibility, and through what mechanism?

Question eleven separates everything. An agency with no answer to "what if nothing changed" will show you progress regardless, because its business model requires it. In a domain where answers naturally drift week to week, a rising line can always be found.

Our own answers to those twelve

For transparency, here is how we answer. We measure 50 to 300 prompts depending on plan, daily or weekly, across ChatGPT, Gemini, Perplexity, AI Overviews and AI Mode. A query with no generated answer is recorded as a coverage gap, never as your absence. We measure position inside the answer, not just the mention.

We write only into dedicated fields we created, never into your customer-facing page, with a check that refuses the write if the product changed in the meantime, and a read-back afterwards. We record a baseline strictly before any write, re-measure at several intervals, and when the gap is too small or the sample too thin, we say so instead of drawing a line.