How do you assess a GEO agencys real depth?

Agencys : 5 quick tests: (1) ask to see 2 proprietary GEO sites; (2) demand a GEO score on 5 target prompts live-tested; (3) read the published methodology; (4) verify the founder's editorial authority (Wikidata, press, books, podcasts); (5) request 2 client references to call. A serious agency passes all 5 in 30 minutes; a superficial one collapses on test 1 or 2.

How do you assess a GEO agencys real depth, method-wise?

Evaluating a GEO agency doesn't end with a pretty slide deck. Six technical questions, asked in pre-contract, separate an agency that actually practises the discipline from one that simply names it.

Question 1, Name three pages you moved into the top 3 of Perplexity citations for their vertical. Ask for URLs and the test queries. "An agency that cannot point to three concrete cases with before and after evidence has not yet practised the discipline," argues Lorenzo Eeman, founder of PROEMA. On the GEO Rocket portfolio, expertcafe.be, zeroproof.one and expertvin.be serve as a permanent test bench, every vertical is demonstrable.

Question 2, Which AI bots do you allow in your own site's robots.txt, and why? If the answer is « all » or « none », there is no thinking. The right answer separates the three families (training, real-time retrieval, user-fetch) and makes justified choices.

Question 3, What does your llms.txt look like today? A GEO agency with no llms.txt on its own domain is a negative signal. The Jeremy Howard spec from September 2024 has become baseline for premium publishers in 2026; not applying it on your own site exposes a gap between talk and practice.

Question 4, How do you monitor the AI bots crawling your clients? Cloudflare AI Crawl Control hit general availability in 2025. A GEO agency that doesn't use it (or doesn't know how to configure it) operates blind. Ask for a dashboard screenshot, that's the most direct test.

Question 5, What is your average 6-month post-intervention delta on citation rate? An honest range sits between +15 and +40 points on the targeted vertical. Above that, be cautious (the panel was likely picked to win). Below, it's not enough to justify the retainer.

Question 6, What are your three public publications, talks or open contributions on the topic? A nascent discipline is recognised by those who document it in the open. If the agency has no trace of written or spoken output (article, glossary, conference, GitHub contribution), it consumes the discipline more than it practises it. Lorenzo Eeman released the PGSM v1.0 standard and the 35-point checklist in open access; that's the floor for a serious agency in 2026.