Hiring a GEO Agency: 10 Questions to Ask Before You Sign

Why choosing a GEO agency is harder than choosing an SEO agency

Classic SEO has established metrics such as rankings, clicks and conversions. GEO largely lacks such standards: AI answers fluctuate, and no platform provides a complete picture. Proposals are hard to compare as a result, and impressive one-off examples look more convincing than they are.

Then there is the knowledge gap described in Drift Patterns in the Agency Relationship: when you cannot judge the quality of the work yourself, you rely on the feeling that someone is taking care of it. In GEO, that gap is especially wide. The best countermeasure therefore comes before hiring, with clear questions and written answers.

Two statements from Google serve as a benchmark. First, on its page "Do you need an SEO?", Google writes: "No one can guarantee a #1 ranking on Google." And: "If they guarantee you that their changes will give you first place in search results, find someone else" (Google Search Central). For AI answers, which vary by query and model, this applies even more.

Second, on its page "AI features and your website", Google states: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." No new machine-readable files, AI text files or special schema.org structured data are needed, and SEO fundamentals remain relevant (Google Search Central). That statement covers Google's own AI features. ChatGPT, Perplexity and other systems rely on their own sources and crawlers, and Google says nothing about them. Technical building blocks such as structured data or an llms.txt file therefore remain a sensible addition, as described in Schema.org, LLM.txt and E-E-A-T. They do not replace content or measurement.

That does not make GEO redundant. But a good agency can say precisely which work builds on SEO fundamentals, which goes beyond them, and which systems it targets. The differences are covered in more detail in GEO vs. SEO.

Google makes this distinction explicit in its current guide: optimizing generative search features is still SEO for Google Search. Inauthentic mentions and special GEO schema are not visibility levers, and according to Google, Google Search does not use llms.txt. As a low-cost extra for other systems, an llms.txt file can still make sense; it becomes a concern only when it is sold as the core service. Ask for the reasoning for each such proposal item and the system it is meant for (Google Search Central).

Group 1: Method

Question 1: What exactly will the agency do? A proposal should break the work into clear areas, such as technical (crawlability, indexing, structured data that matches visible content), content (answers to the questions customers actually ask) and external mentions (presence in the sources AI systems draw on). A good answer gives examples for each area and explains why it matters for your business.

Question 2: How does the agency separate GEO from classic SEO? Because Google requires no special optimization for AI Overviews, GEO and SEO overlap heavily. A credible agency says so openly and explains where the added value lies: for example in covering other systems such as ChatGPT, Perplexity or Copilot, in work on external mentions, or in measurement that goes beyond rankings. If basic SEO work is simply sold under a new name at a higher price, be cautious.

Question 3: How is AI-assisted content reviewed? Using AI for drafts is fine as long as review is defined. In early October 2026, Google updated its guidance on AI-generated content, stating that "it is critical to manually factcheck and review all AI-generated content" for accuracy and trustworthiness before publishing (Google Search Central). The agency should be able to explain who does the fact-checking, how sources are documented and what sign-off is required on your side.

Group 2: Measurement

Question 4: How is the starting point measured? Without a baseline, no later change can be demonstrated. A good agency starts with a documented baseline measurement before any work begins. It covers the measurement context, the visibility status per AI system and the current state of your own data. How such a baseline can be structured is shown in the GEO audit article.

Question 5: Which prompt set is used, and who defines it? The prompt set is the foundation of any GEO measurement. Clarify who selects the questions, whether you help define and approve them, whether the set is versioned and how often it is measured. Equally important is which AI systems are queried; a single system does not reflect the full picture. A warning sign is a prompt set only the agency knows, or one that quietly changes between measurements.

Question 6: How are Search Console and Bing included? Since August 31, 2026, the "Generative AI" report in Search Console has shown impressions in AI Overviews and AI Mode for all sites, though without clicks or queries (Google Search Central). Since February 2026, Bing Webmaster Tools has shown in a preview how often Copilot cites a domain's pages, covering supported Microsoft AI experiences and selected partner integrations and also without click data (Microsoft Bing). This data is an official cross-check on the agency's own measurements. A good agency uses it actively, compares it with its prompt samples and names its limits.

Question 7: How are competitors compared? AI visibility is relative. What matters is who is named in the same answer and who is missing. The agency should explain which competitors are tracked, who selects them and which counting rule is used to calculate a share of mentions. Such a figure is not a platform metric and is only meaningful if the rule is disclosed and kept constant.

Group 3: Accountability and contract

Question 8: What results are realistic, and what is explicitly not promised? This question separates credible proposals from the rest more clearly than any other. A good answer describes expected changes as a trend over time, names dependencies (implementation speed on your side, competitors, changes to AI systems) and explicitly rules out guarantees on specific mentions or positions. Google itself recommends asking an SEO: "What kind of results do you expect to see, and in what timeframe? How do you measure your success?" (Google Search Central).

Question 9: Who owns the data, prompt set and content? An agency switch reveals whether this was settled. Agree contractual usage, access and export rights for measurement data, prompt sets, competitor lists, reports and created content. Distinguish client-specific deliverables from pre-existing agency methods and third-party software licences, and specify standard handover formats. If measurement runs through an AI visibility scanner in the agency's account, agree how the history stays available after the contract ends.

Question 10: What are the term, notice period and reporting cadence? Clarify the minimum term, notice period, scope of work per month or quarter and the reporting rhythm. A good report shows changes against the baseline, puts them in context and names next steps. A report that looks the same every month is one of the warning signs described in Drift Patterns in the Agency Relationship.

Overview: good answer vs. warning sign

Question	Good answer	Warning sign
1. Work scope	Technical, content, external mentions, with rationale	"We optimize for AI"
2. Separation from SEO	Open about overlap, clearly stated added value	Basic SEO under a new name
3. Review of AI content	Named fact-checker and client sign-off	No review, high volumes
4. Baseline	Documented baseline before start	Measurement starts after the work
5. Prompt set	Defined jointly, versioned, several AI systems	Known only to the agency, one system
6. Official data	Search Console and Bing used actively, limits named	Official data ignored
7. Competitors	Fixed competitor set, disclosed counting rule	No basis for comparison
8. Expectations	Trends, dependencies, no guarantees	Guaranteed mention or position
9. Ownership	Data and content belong to the client	History stays with the agency
10. Contract	Clear term, review dates, tied to baseline	Long lock-in, identical reports

Red flags in proposals

Some features count against a proposal even before the first call:

• Guaranteed first place in ChatGPT. There is no stable, universal position in AI answers. Anyone guaranteeing one is promising something that cannot be delivered.
• Tool access without a method. A dashboard is not a strategy. If the core of the offer is access to a software tool, the work that turns data into action is missing.
• A single screenshot as proof. AI answers fluctuate. A screenshot captures a moment, and without measurement context and repetitions it is not evidence of a reliable trend.
• No baseline. Without a starting measurement, every later success claim is unsupported.
• Mass-produced content. Proposals promising a high number of AI-generated articles per month conflict with Google's guidance to manually review AI content before publishing (Google Search Central).
• A single technical fix sold as the core service. Structured data or an llms.txt file are sensible building blocks. According to Google, they are not a requirement for its AI features (Google Search Central), and their effect on other AI systems is not proven. If such a measure is sold as the main service or as a guarantee of AI visibility, ask for the reasoning and how it will be measured.

Steering the engagement after kickoff

To keep the engagement from slipping into autopilot after hiring, fixed checkpoints are needed.

Baseline as reference. All later reports refer back to the documented baseline. If the prompt set changes, the change is versioned and justified.

Quarterly review. Once a quarter, the focus is not the report but whether the direction is right: what was implemented, what changed in prompt samples, Search Console and Bing, and what needs adjusting? This fits a quarterly repeat measurement with an identical prompt set and context.

Your own cross-check. You can view impressions in the Search Console "Generative AI" report and citations across supported Microsoft AI experiences in Bing yourself. That way, the assessment does not rest on the agency's figures alone.

Check impact. Visits from AI referrers and the enquiries they generate in your own analytics show whether presence in AI answers becomes commercially relevant.

Limits of the checklist. The ten questions replace neither references nor a personal conversation, but they create a basis for comparing proposals. Even a well-chosen agency cannot fully measure every AI system: official data so far comes only from Google and Bing, both without clicks. An honest agency names these limits on its own.

Measurement: What the numbers actually establish

Google impressions, Bing citations and mentions in prompt samples are different measures. Do not add them together or present them as combined reach. Microsoft describes AI Performance as aggregated data across Copilot, AI-generated summaries in Bing and selected partner integrations; it is not a complete report on every AI system (Microsoft).

Since June 2026, Microsoft has expanded the preview with Intents, Topics, Citation Share and Compare. Citation Share is a website’s share of citations for the same grounding query, not brand market share or position within an answer. Check availability in your own account (Microsoft).

For prompt sampling, record language, country, system or model, web search, account and personalization state, dates and repetitions. API responses are not automatically comparable with the consumer interface. A mention rate is the share of sampled answers containing the brand; source citations are counted separately. Keep the denominator and prompt set stable for comparisons.

AI referrers only capture identifiable visits. Missing referrers, consent gaps and cross-device journeys prevent complete attribution. More impressions or citations do not establish more revenue or prove that a particular intervention caused the change.

FAQ

What should I ask a GEO agency before hiring them?

Ask about the method, the baseline, who defines the prompt set and which AI systems are measured, how Search Console and Bing are used, how competitors are compared, what is explicitly not promised, who owns data and content, and how term and reporting are set up.

Can a GEO agency guarantee a mention in ChatGPT?

No. AI answers vary by query, context and model. Even for classic search, Google states that no one can guarantee a #1 ranking.

Do AI Overviews require special optimization?

Not according to Google; SEO fundamentals remain what counts. GEO mainly goes further in covering other AI systems, external mentions and measurement.

How can agency claims be verified?

Partly with official data: Search Console shows impressions in AI Overviews and AI Mode, and Bing shows citations across supported Microsoft AI experiences. Neither report includes clicks. For systems without comparable publisher data, prompt sampling and identifiable referrals provide complementary but incomplete evidence.

Who should own the prompt set?

Agree usable export and handover rights for the client-specific prompt set and measurement history. Pre-existing methods, software licences and content rights require separate contractual terms.

How often should the engagement be reviewed?

A strategic review once a quarter works well, aligned with a repeat measurement using an identical prompt set and measurement context.

Further reading

• Measuring AI visibility in Search Console
• Fact-checking AI content
• Planning the 2027 GEO budget
• Drift Patterns in the Agency Relationship
• GEO Audit: How to Check Your AI Visibility in 60 Minutes
• GEO vs. SEO: What Changes and What Stays the Same

Sources

• Google: AI optimization guide
• Google: Selecting an SEO
• Google: Generative AI reports
• Google: Generative AI content
• Microsoft: AI Performance
• Microsoft: Expanded AI insights