Choosing a GEO agency: 9 RFP questions and five red flags
A report showed "AI search visibility up 240%." Your leadership was satisfied until procurement asked, "How many orders actually came from that?" No one answered. The GEO market is growing fast and remains disorganized. Here are nine RFP questions and five red flags to include in vendor selection, straightforward approaches to look past the metrics and find what actually moves your business.
By
Tenten AI FDM 團隊
前線部署行銷
Published
March 26, 2026
Read time
6 分鐘
Last quarter, a B2B software company asked us to review a contract. They'd signed with a GEO agency six months prior, not a discount arrangement. Each month, a polished report arrived: "Your brand's AI search visibility is up 240%." The leadership team was satisfied until the procurement director asked, "How many of those deals came from customers who said they found us on ChatGPT?" The room went quiet.
Looking at the report, the issue appeared on page three. The "240% visibility growth" came from running 30 keywords through Perplexity and calculating their own weighted score. No control group. No competitive baseline. No disclosed prompt. The number was shaped to impress, not to inform decisions.
Choosing a GEO agency: How do you prove this works?
The GEO market (Generative Engine Optimization) is growing and disorganized. Many teams moved from SEO last quarter and simply renamed themselves GEO experts on their pitch deck. The first thing to assess isn't how polished their case studies look, but whether they'll tell you straight: "Here's what we can control, and here's what we can't."
AI search differs fundamentally from Google rankings: you cannot guarantee placement. The same prompt might reference you today and a competitor tomorrow, depending on model versions, what's indexed, and timing. A vendor promising "guaranteed ChatGPT recommendations within three months" either doesn't understand the landscape or plans to show you misleading reports.
9 Questions to Ask in an RFP
Use these nine questions in your RFP. Agencies confident in their work will answer directly. Those without substance will deflect.
| # | Question | What You Want to Hear |
|---|---|---|
| 1 | How do you define "success"? What business metric is it tied to? | Real traffic, inquiries, and revenue, not "visibility scores" |
| 2 | Which engines and prompts do you measure against? Will you share them? | Full prompt list and engine versions; methodology is reproducible |
| 3 | Do you have competitive benchmarks and before/after baselines? | Improvement measured against your competitors in the same window |
| 4 | What do you actually do, or do you just consult? | Hands-on work: content rewrites, structured data, entity building, not just advice |
| 5 | Who writes the content? Do you know my industry? | Real editors who understand your sector, not AI bulk copy |
| 6 | How do you handle the technical foundations, structured data, crawlability? | They can actually work with your engineering team, not just hand over a Word doc |
| 7 | How often do you report? Can I see the raw query logs? | Auditable raw data, not just screenshots |
| 8 | If we see nothing after three months, what's the exit? | Clear checkpoints and a real exit clause; no multi-year lock-in |
| 9 | Tell me about a project that failed or where you hit limits. | If they can answer this honestly, they've actually done real work |
Question 9 separates experience from marketing hype. Teams that have done dozens of campaigns have hit real obstacles. If they can't identify a project that underperformed, they're either new or being dishonest.
Red flags in inflated reports
Most projects we've reviewed in the past year shared the same core issue: measurement designed to obscure rather than actual execution problems. Watch for these red flags:
Only a top-line percentage with no detail. All visibility compressed into one number, but they won't say which prompts they used, how many iterations, or when they measured. Numbers like this can be manipulated infinitely. They're meaningless.
No competitive baseline. Your visibility climbed, but so did your competitors'. That's likely the model indexing your entire category, not the agency's work.
They're claiming credit for branded keywords. Reports highlight "What is Tenten?", searches from people who already know you. Real progress shows in non-branded searches like "Taiwan GEO consultant recommendations," where you had no visibility and had to earn the mention.
Numbers disconnected from business results. They never ask whether visibility became traffic, leads, or deals. GEO isn't a scoring game. If it doesn't move your pipeline, it's performance theater.
Long-term contracts with no exit clause. Confident agencies include review gates. If they won't accept a year-end checkpoint before locking you in, they're betting the contract holds you, not the results.
The litmus test
Picking a GEO agency isn't different from hiring any engineering vendor: look for someone who'll own their limitations. AI search is unpredictable by nature. Honest work means reproducible measurements, auditable raw data, and targets tied to business outcomes. Not dashboards that look better each week.
We hold our GEO work to this standard: every number traces to the exact prompt and timestamp. The only metric that counts is whether it brought real inquiries. Visibility is pleasant to see, but it only matters if the AI referenced you and a customer actually found you as a result.
One stuck workflow
is enough to begin
Tell us what the team does today, where it breaks down, and what a better working day should look like.