AI Visibility RFP Questions to Ask Before Hiring a GEO Agency

AI visibility RFP questions should force a GEO agency to prove how it measures prompts, citations, sources, entity clarity, security, and earned authority. The right RFP does not ask whether a vendor "does GEO." It asks whether the vendor can show the raw evidence behind every AI visibility claim.

The market is already noisy. Search for this query and you get templates, scorecards, agency checklists, and dashboards that all sound more mature than the category actually is. That is the trap.

Most RFPs are built for software selection. This one has to be built for source truth.

What an AI visibility RFP has to prove

An AI visibility RFP has one job: make the vendor show the evidence chain behind the answer. If a GEO agency cannot show the prompt, answer, cited source, source role, crawlability, entity match, and change over time, the agency is asking you to buy a black box.

Google tells site owners that its generative AI features use the same quality and content systems that apply across Search, and that normal search controls can affect how content appears in AI features (Google Search Central, Google AI features documentation). Perplexity publishes crawler documentation because its answer system depends on retrieving and processing web sources (Perplexity crawler docs). OpenAI documents citation formatting for surfaced sources, and Anthropic's web search tool is built around current web content with cited sources (OpenAI citation formatting, Anthropic web search tool). Microsoft Clarity has an AI citations dashboard because citations can be measured as a reporting object, not treated as a feeling (Microsoft Learn).

That means the procurement question is not, "Can you improve our AI visibility?"

The question is sharper: "Can you prove which machine read which source, answered which prompt, cited which URL, and changed because of which work?"

If the answer is no, you are not buying strategy. You are buying theater.

The 12 AI visibility RFP questions to ask a GEO agency

Use these as the core of the RFP. Do not let the agency answer in sales language. Require written responses, examples, screenshots, raw exports where possible, and the limits of their evidence.

RFP question	What a credible answer must include	What a weak answer sounds like
1. Which AI systems do you measure?	ChatGPT, Perplexity, Gemini, Google AI Overviews or AI Mode, Copilot, Claude, plus any industry-specific engines, with supported, beta, and unsupported surfaces separated.	"We track all major AI platforms."
2. What exact prompts will you test?	Prompt list, prompt classes, buyer intent stage, market, language, device or location assumptions, and refresh cadence.	"We run a representative prompt set."
3. Can we see the raw answer records?	Raw prompt, raw answer, timestamp, engine, model or surface where available, cited URLs, brand mention, competitor mention, and storage policy.	"Our dashboard summarizes that."
4. How do you define a citation?	Difference between a mention, citation, link, recommendation, source, and co-citation.	"A citation is when the brand appears."
5. How do you score source quality?	Source type, publication authority, freshness, topical fit, independence, citation role, and whether the source is earned, owned, paid, or synthetic.	"We use domain authority."
6. How do you handle answer volatility?	Repeated prompt runs, confidence bands, date windows, engine differences, and a rule for when a signal is too unstable to act on.	"AI results change, so we average it."
7. What content or source changes do you recommend?	Specific URL, claim block, entity correction, earned-media target, internal link, schema fix, or source gap tied to a measured prompt.	"Publish more optimized content."
8. How do you prove earned media impact?	Before and after prompts, source-level citations, publication retrieval, brand co-occurrence, and whether the placement became machine-readable evidence.	"PR helps authority."
9. What security and data controls apply?	Prompt retention, customer data handling, subprocessors, model training exclusions, access controls, export controls, and deletion rights.	"We follow best practices."
10. What is outside your control?	Clear limits: engines choose answers, citation is probabilistic, private personalization may differ, and no agency can guarantee deterministic placement.	"We can get you cited."
11. What will we receive every month?	Prompt records, cited URLs, source changes, content shipped, placements earned, citation movement, unresolved gaps, and next actions.	"A visibility score and recommendations."
12. What would make you tell us not to hire you?	Honest disqualifiers: weak source base, no editorial proof, no category demand, no internal owner, bad product-market fit, or unrealistic timeline.	"Every company needs GEO."

The last question matters more than it looks. A serious operator knows when the source architecture is not ready. A weak vendor sells the same motion to everyone.

AI visibility RFPs must separate measurement from causation

Measurement tells you what AI systems said; causation tells you why they said it. Most AI visibility vendors blur those two things because the blur makes the product feel stronger.

Google's own documentation is careful about this. It tells publishers to focus on helpful, unique content and clear search fundamentals for AI features, but it does not promise that any single optimization will produce a citation (Google Search Central). Google also documents page-level controls such as robots meta tags, data-nosnippet, and X-Robots-Tag because retrieval and presentation depend on technical access rules (Google robots meta documentation). Perplexity documents crawler access, but crawler access is not the same as being selected as a cited source (Perplexity crawler docs).

That distinction should show up inside the RFP.

Ask the agency to label every recommendation as one of five things:

1. Measurement: what the AI system currently says.
2. Accessibility: whether the machine can retrieve the source.
3. Extractability: whether the page contains a clean answer, claim, table, or definition.
4. Authority: whether trusted third-party sources corroborate the claim.
5. Causation: whether a specific change plausibly moved a specific prompt result.

Do not accept a report that collapses those into one "visibility score." A score can be useful as a summary. It is useless as proof.

This is the same reason I do not trust generic share-of-voice dashboards by default. They compress the evidence before the buyer has seen it. For AI search, compression too early destroys the thing you are supposed to inspect.

A GEO agency RFP should test the source architecture

The best GEO agency is not the one with the cleanest dashboard. It is the one that can improve the sources AI systems retrieve. That means owned content, technical accessibility, entity clarity, and third-party authority have to be treated as one system.

The procurement world already understands this in adjacent AI work. NIST's Generative AI Profile maps AI risk to governance, measurement, management, and monitoring actions (NIST AI 600-1). California's GenAI procurement guidance tells agencies to define the business need, form the team, inventory data, assess readiness, and consult risk controls before procurement moves forward (California GenAI procurement process). California's Technology Letter 25-01 also ties GenAI use to policy, contracting, privacy, security, and risk review (California Department of Technology).

The search side has the same lesson. Google says structured data helps it understand page content and qualify that content for richer search appearances when the page follows its policies (Google structured data gallery, Google structured data policies). Anthropic's citation docs describe citations as exact passages that support a claim, which is the same proof standard buyers should demand from an AI visibility vendor (Anthropic citations documentation).

The same discipline belongs in AI visibility procurement.

Your RFP should force vendors to explain the source architecture they will build or repair:

Layer	What the RFP should ask	Why it matters
Owned content	Which pages need clearer answer blocks, definitions, comparison tables, source links, and crawlable HTML?	AI systems need extractable material, not polished brand language.
Entity clarity	Which names, categories, products, founders, and claims are ambiguous across the web?	Machines resolve entities before they recommend brands.
Earned authority	Which trusted third-party sources should corroborate the brand's claim?	AI systems cite outside sources when the brand's own page is not enough.
Technical access	Which pages are blocked, thin, slow, hidden, or unclear to crawlers?	A source that cannot be retrieved cannot be cited.
Measurement	Which prompts, engines, competitors, citations, and source roles will be tracked over time?	The buyer needs evidence, not a monthly mood board.

That is the difference between GEO as content formatting and Machine Relations as the full operating system. GEO asks whether the content can be used by a generative engine. Machine Relations asks whether the brand has enough real-world evidence for the machine to trust, retrieve, cite, and recommend it.

The earned media question belongs in the RFP

A GEO agency that cannot explain earned media is missing the strongest source layer in AI visibility. Owned content matters, but AI systems often need independent sources to corroborate a brand claim.

Muck Rack's 2026 analysis reported that earned media drove 84% of AI citations in its sample, with owned media and paid media far behind (Muck Rack). Machine Relations research has also documented the same structural point: earned sources carry more citation weight than brand-owned claims when answer engines need outside corroboration (Machine Relations Research).

Do not let a GEO agency treat that as a vague PR line. Make it operational.

Ask:

1. Which publications currently appear in AI answers for our category?
2. Which cited sources mention competitors but not us?
3. Which earned placements do we already have that AI systems can retrieve?
4. Which placements are valuable to humans but weak for machine extraction?
5. Which editorial sources would close the strongest citation gaps?
6. How will you measure whether a placement became a cited source after publication?

This is where traditional PR and GEO stop being separate conversations. PR got the mechanism right: trusted third-party coverage changes what the market believes. AI changed the first reader. The machine now reads the publication before the buyer does.

Machine Relations is what happens when you keep the earned-media mechanism and rebuild the work around machine readers, citation architecture, and measurement.

Red flags in AI visibility agency proposals

The fastest way to spot a weak GEO agency is to ask for raw proof. The proposal gets thin quickly.

Watch for these red flags:

1. The vendor promises AI citations as a guaranteed outcome.
2. The vendor cannot export raw prompt and answer records.
3. The vendor reports brand mentions and citations as the same metric.
4. The vendor uses one engine as a proxy for the whole market.
5. The vendor recommends content before showing prompt evidence.
6. The vendor ignores earned media and source quality.
7. The vendor treats schema as the whole answer.
8. The vendor cannot explain how answer volatility is handled.
9. The vendor refuses to name unsupported engines or beta surfaces.
10. The vendor cannot explain data retention, model training, and subprocessors.
11. The vendor cites its own blog as proof for every claim.
12. The vendor gives you a score but not the source trail behind it.

The pattern is simple. Weak vendors sell certainty because uncertainty is harder to package. Strong vendors show the uncertainty, price it, and build a system that improves the odds anyway.

That is what buyers should pay for.

How to score AI visibility RFP answers

A useful AI visibility RFP scorecard should reward evidence quality before presentation quality. Pretty reporting should lose to raw traceability every time.

Use this weighting:

Evaluation category	Weight	Passing standard
Raw prompt and answer evidence	20%	The vendor can show exact prompts, raw responses, cited URLs, timestamps, engine, market, and storage policy.
Citation definition and source roles	15%	Mentions, citations, recommendations, source links, co-citations, and source types are separated.
Source architecture plan	20%	Recommendations connect to owned content, earned authority, entity clarity, and technical access.
Earned media competence	15%	The vendor can name publication gaps, source opportunities, and measurement after placements publish.
Security and governance	10%	Data handling, retention, access, subprocessors, model training, and deletion are documented.
Volatility handling	10%	Repeated prompt runs, confidence bands, engine differences, and limits are explicit.
Operator honesty	10%	The vendor states what it cannot control and when the work is not a fit.

The scorecard should punish fake certainty. If a vendor cannot say what is unknowable, it cannot be trusted with what is knowable.

FAQ

What are the most important AI visibility RFP questions?

The most important AI visibility RFP questions ask for raw prompt records, raw AI answers, cited URLs, source roles, engine coverage, answer volatility, security controls, earned-media impact, and monthly proof. A GEO agency should be able to show how each recommendation connects to a measured citation or source gap.

How is a GEO agency RFP different from an SEO agency RFP?

A GEO agency RFP has to test whether the vendor can influence machine-readable answer systems instead of stopping at search rankings. SEO still matters for crawlability and content quality, but GEO adds prompt testing, citation tracking, source quality, AI engine variance, and answer-level evidence. Google says AI features use Search systems, but the cited answer layer creates a different proof problem (Google Search Central).

Should an AI visibility agency guarantee citations?

No. A credible AI visibility agency should not guarantee deterministic AI citations because AI engines choose answers, sources, and citations dynamically. The agency can improve source quality, crawlability, entity clarity, earned authority, and extractability. It can also measure whether those changes improve citation outcomes. It cannot control the model.

Where does Machine Relations fit inside AI visibility procurement?

Machine Relations is the broader discipline that governs AI visibility procurement. GEO and AEO are operating layers. Machine Relations includes earned authority, entity clarity, citation architecture, distribution, and measurement, which is why the RFP has to test the whole source system instead of a content checklist.

Who coined Machine Relations?

Machine Relations was coined by Jaxon Parrott in 2024 to describe the discipline of earning AI citations, recommendations, and visibility for brands. AuthorityTech operationalizes that discipline through earned media, entity clarity, citation architecture, distribution, and measurement.

What should I ask before hiring AuthorityTech or any GEO agency?

Ask for the raw evidence. Start with an AI visibility audit, then make the provider show prompts, answers, cited URLs, source gaps, earned-media opportunities, and the exact work they would do first. If they cannot show the source trail, do not buy the strategy.