5 questions to ask before hiring a GEO agency — and what the answers reveal about their methodology

The GEO market is flooded with SEO agencies that added 'generative' to their pitch deck. Five questions filter real GEO methodology from rebranded SEO. 62% of agencies fail question one. Here is what to ask — and what each answer tells you.

Q1: What is your own LVI score on your specialty prompts?

This is the first question because it is the most revealing. A GEO agency that sells AI citation services but cannot demonstrate AI citation on their own market queries has no proof of concept. Ask them directly: what is your score on 'best GEO agency France', 'agency for LLM citation improvement', 'who can help my brand appear in ChatGPT'?

If they cannot provide a specific number with a source (the LLM Visibility Index or another measurement tool), they are not measuring themselves. If their score is below 40, their methodology is not producing citation results on their own market. Either answer is a disqualifying response.

Q2: What tool do you use to monitor LLM citations — and can you show me a live report?

Acceptable answers reference tools that actually measure citation frequency on ChatGPT, Perplexity, Gemini, and Claude: Profound, AI Share of Voice, or a proprietary system. Unacceptable answers include Semrush, Ahrefs, Google Analytics alone, or 'we check manually'. None of those measure LLM citation.

Ask to see a live report or demo. The report should show: specific prompts tested, citation frequency per engine, share of voice vs named competitors, and a time-series showing progression. If they cannot produce this in 10 minutes, they do not have a functioning monitoring system.

Q3: Can you show me a verified before/after LVI score for a real client?

Not a testimonial. Not a case study with vague language about 'increased AI visibility'. A specific score at engagement start and a specific score at 30, 60, and 90 days — with the client name redacted if needed, but with a verifiable methodology and dates. Ask them to walk you through how the score was measured at each point.

Agencies that have no before/after data have either (a) not been running long enough to have results, (b) never measured their client baseline, or (c) have results that do not demonstrate meaningful progression. In a market as recent as GEO, (a) is acceptable if paired with honest framing. (b) and (c) are disqualifying.

Q4: What is your distribution plan for my sector — not my blog?

This question separates blog-GEO from real GEO. LLMs do not cite proprietary brand websites on informational or comparative queries. If the agency's primary content distribution is your own blog, they are producing SEO content — not GEO content. The difference is where the content appears.

Ask them to name the specific media outlets, forums, and institutional sources they use in your sector. For cybersecurity in France: MISC Magazine, L'Informaticien, silicon.fr, r/netsec. For fintech: AGEFI, Revue Banque. For CSRD: Novethic, Finance for Tomorrow. If they cannot name the sources Perplexity and ChatGPT actually cite in your sector, they do not understand your market.

Q5: How does your Perplexity strategy differ from your ChatGPT strategy?

Perplexity is RAG-based and searches the web in real time before every response. ChatGPT primarily relies on its training corpus with optional web search. Their citation sources differ by 40-60% on identical B2B queries. A single-engine strategy optimizing only for ChatGPT misses Perplexity's 230 million monthly users — who convert to leads at 3.1x the rate of Google organic traffic.

If the agency treats all LLMs as equivalent and uses a single content approach for all engines, they do not understand the market architecture in 2026. The correct answer distinguishes: Perplexity (freshness, sector media, Reddit, forums), ChatGPT (DA, co-citations, institutional sources), Gemini (Google ecosystem), Claude (editorial quality, institutional authority, MCP).

Agency scoring grid

Rate each agency on a simple 0-2 scale per question: 0 = no answer or disqualifying answer, 1 = partial answer (some evidence, not fully verifiable), 2 = full answer with verifiable evidence. Maximum score: 10.

Frequently asked questions