How to choose a GEO/AI search agency: 7 evaluation criteria and 15 due diligence questions

Late last year, an industrial precision parts manufacturer brought us in to handle a crisis. They'd hired an agency the year before that hung out a 'GEO optimization' shingle. Every month, the agency delivered 30 blog posts and a polished keyword report, all with 'AI friendly' headlines. After ten months of payments, the problem surfaced. When their procurement managers asked ChatGPT or Perplexity 'Who are the precision parts suppliers in Taiwan?', the client's name never appeared in the answers. Content production hit its targets. AI citations: zero.

This is the most common trap in the GEO market. Choosing a GEO agency isn't about buying content volume; it's about buying the ability to get cited by AI. Can your brand actually show up in answers on ChatGPT, Perplexity, or Google AI Overviews, with verifiable citation screenshots to prove it? If you confuse these two, you'll end up like that manufacturer. Ten months and a stack of articles nobody reads, that machines won't cite either.

Choosing a GEO agency: check citation track record first, not content volume

The quickest way to gauge whether a GEO agency is worth signing is to ask for their track record of citations: which current clients' brands have been cited by which AI engines, answering which specific questions, with screenshot proof and the URLs of cited pages.

Content mills are everywhere. One prompt generates ten posts a day. But whether an AI engine will cite you depends on whether your content has machine-readable structure: clear definitions, concrete data points, FAQs, schema markup, credible sources. Any agency that just pumps out articles and can't show you a single screenshot of real AI citations is selling you 2015-era SEO content farming with an AI wrapper. This is what your procurement team needs to hold.

Seven evaluation criteria

If they can't provide concrete answers on the first two items (citation track record and measurement method), cut them loose, no matter how polished the rest of the pitch is.

15-question due diligence checklist: just ask these

Bring this checklist into the pitch meeting and work through it item by item. If they can't answer, or just pivot away, you know they're selling volume, not citations.

1. Show me: Which of your current clients' brands have been cited by ChatGPT or Perplexity? Give me answer screenshots and the URLs that were cited.
2. How do you measure success? Which engines do you track, and how often?
3. How do you define and calculate 'citation rate' or 'AI share of voice'? Can you show me a real sample report?
4. If in three months my brand still isn't appearing in AI answers, what's your remediation process and exit clause?
5. Who writes the content? Humans, AI, or hybrid? What's your fact-check process?
6. Will you touch the technical layer of my site? Who handles schema, llms.txt, structured data?
7. For my industry, can you list 10 actual questions your buyers would ask an AI?
8. What projects have you done in my industry or similar ones? Can you give me a reference I can actually call?
9. After content goes live, how do you verify it's machine-readable and citable, not just pretty for humans?
10. How do you connect AI mentions back to my actual business results: inquiries, pipeline, deals? How's attribution structured?
11. What does this contract actually deliver each month? Number of pieces, technical changes, reports, in writing, in detail.
12. Can I see the workspace and progress in real time, or is it just a PDF once a month?
13. When an AI engine changes its algorithm or policy, how quickly do you find out, and how quickly do you adapt?
14. Who owns the content and data? Can I take everything with me when the contract ends?
15. Who's actually doing this work: senior people or the cheapest subcontractor they can find?

One question to ask before signing

When we do FDM (front-line deployment marketing), we don't start with 'here's how many articles you'll get.' We map out the actual questions your buyers take to AI, then engineer the structure, data, and schema needed to get cited, and track every citation back to a screenshot. The number of articles we deliver has never been our metric. What matters is whether your brand actually shows up in that AI answer.