How should a business choose an AI visibility agency?
If you're comparing GEO, AEO or AI SEO consultants, ask how they investigate real buyer questions, verify source evidence and report useful outcomes—not merely whether they display an impressive score.
How do you choose an AI visibility agency?
Choose an AI visibility agency by reviewing its evidence-gathering method, relevant technical and editorial capabilities, measurable deliverables, transparent limitations and ability to turn findings into useful changes. Ask for a defined buyer-question sample, transparent citation checks, a written action plan and a reporting method that does not confuse mentions with recommendations or actual customers.
An agency may market its services as AI visibility, generative engine optimization (GEO), answer engine optimization (AEO), LLM SEO or AI search optimization. Those names can overlap. Do not select a provider solely because it uses your preferred label. Establish which problems it will diagnose, which changes it will make and what evidence it will use to evaluate them.
This guide is a vendor-neutral selection framework. MentionGauge offers manual AI visibility audits, but the same questions should be applied to MentionGauge, to competing agencies and to internal teams. The scoring exercise below is illustrative, not a ranking or endorsement of any firm.
Which kind of agency or specialist do you need?
First distinguish a visibility diagnosis from the work required afterward. A monitoring subscription may be premature if nobody has checked the site's accessibility or defined meaningful buyer questions. Conversely, a one-off audit may be insufficient when your team cannot implement the findings.
| Your main need | Capability to look for | Useful proof to request |
|---|---|---|
| Understand why your brand is missing | Manual AI visibility audit | Sample prompt panel, answer evidence, source checks and findings prioritization |
| Improve answers to buyer questions | AEO and editorial information architecture | Question-to-page map, editorial before/after examples and validation rules |
| Strengthen generative answer relevance | GEO and source credibility work | Citation/source analysis and documented recommendations with caveats |
| Fix technical or content problems | AI search implementation and technical SEO | Crawl/render/canonical findings, change log, release QA and site ownership |
| Watch change over time | AI visibility monitoring | Repeated-prompt baseline, sample definitions, reports and alert review |
A good provider can explain where these roles intersect without claiming that one proprietary trick solves them all. Compare our distinct AEO, GEO, implementation and monitoring descriptions for examples of how scopes can be separated.
What should an AI visibility audit actually test?
Ask how the agency selects questions from real sales calls, search queries, support tickets, product categories and competitive research. A useful panel includes nonbranded buyer questions and selected branded controls, not only variations of the client's company name. The agency should explain the target audience, language, geographic relevance, chosen AI platforms, test mode, repeat count and any personalization limitations.
A written example question such as “Which vendors offer enterprise onboarding software for regulated finance teams?” is more reproducible than “Check our AI rankings.” The exact question and conditions matter. Ask whether results are recorded with timestamps and whether failed or incomplete runs are kept in the dataset rather than quietly discarded.
How should the agency evaluate citations and recommendations?
Request separate classifications for brand mentions, provider recommendations, own-domain citations, third-party source citations, competitor mentions and measurable referrals. A webpage linked as background information is not necessarily the company being recommended. A screenshot of one favorable response does not establish repeatability or business impact.
Researchers should open cited URLs, confirm that they exist, distinguish an original source from a derivative summary and record when a platform shows no usable citation. They should separate what was directly observed from a hypothesis about why the model answered that way. A vendor that claims direct insight into a proprietary model's hidden retrieval process should be able to substantiate that claim rather than merely asserting it.
Which deliverables should be included in the proposal?
Before signing, ask for a written scope that names the platforms, questions, number of runs, pages, source-review depth, reporting frequency, implementation tasks, stakeholder inputs and exclusions. The precise package can vary, but a decision-useful engagement usually has recognizable evidence artifacts rather than only a proprietary score.
- Research plan: audience segments, intent groups, selected prompts, geography, languages, platform versions or modes where observable, and sampling caveats.
- Observation record: dated answers, URLs, competitors, mention/recommendation labels, screenshots when permitted and reviewer notes.
- Verified findings: accessibility, canonicalization, source errors, outdated claims, missing important answers and distinctions between confirmed defects and plausible explanations.
- Prioritized recommendations: actions tied to owners, effort, expected mechanism, dependencies, verification method and practical limits.
- Implementation evidence: if changes are in scope, changed URLs or files, approvals, QA results and release notes.
- Follow-up plan: a comparable repeated prompt panel plus business metrics, with warnings against treating visibility fluctuations as causal revenue proof.
What should a sample report look like?
A report should let another reviewer reproduce the reasoning. For example, an illustrative finding could say: “In six completed observations of a defined procurement question, Brand A was mentioned twice and recommended once; two displayed citations referenced outdated product information. We verified one broken internal link and one conflicting feature description. Update the affected pages, then retest the same question panel.” Those counts are fictional and used only to show the level of detail to request.
The sample should also contain a record of what remains unknown. An honest report might say that a competitor appeared more frequently in the observed sample but that the available data cannot prove the precise reason or predict next month's answers.
What technical SEO and content capabilities should you verify?
A competent agency should understand ordinary web foundations: robots restrictions, canonical URLs, renderable HTML, internal linking, duplicate content, sitemap consistency, crawlable service pages and accessible mobile content. It should be able to improve answer quality without creating dozens of thin pages for slight keyword variations. Structured data must describe visible, accurate content, not imaginary AI placements.
Ask how the provider differentiates informational guides, implementation pages, paid audits and commercial service pages. This matters because generic GEO, AEO and AI SEO pages often compete for overlapping intents. A sound content architecture assigns one primary owner to each significant search intent and uses supporting pages for distinct questions. Google advises publishers to follow people-first content guidance and its AI features documentation rather than relying on an invented special ranking recipe.
Use a practical vendor evaluation scorecard
Score each category from 0 to 3: 0 = no evidence, 1 = vague claim, 2 = clear method with an example, 3 = reproducible proof appropriate to the proposed scope. This is a suggested procurement aid, not a scientific score or a validated ranking of agencies. Adjust category weights to your business needs.
| Category | Evidence to request | Score |
|---|---|---|
| Relevant client and buyer research | Specific intent clusters, audience fit and documented prompt selection | 0–3 |
| Source and answer verification | Linked evidence, reproducible observations and clear labels | 0–3 |
| Technical/content competence | Actual audit examples, implementation workflow and quality checks | 0–3 |
| Transparency and limitations | Conditions, assumptions, uncertainty and honest claims | 0–3 |
| Deliverables and ownership | Specific outputs, responsibilities, acceptance criteria and access control | 0–3 |
| Measurement and commercial alignment | Comparable baseline, outcome definitions and reporting cadence | 0–3 |
Maximum illustrative score: 18. An agency with a higher total is not automatically superior if it lacks a required technical skill or cannot work within your compliance rules. Treat failed mandatory checks—such as fabricated case studies, ownership ambiguity or guaranteed placement claims—as disqualifying regardless of score.
Which questions should you ask in a discovery call?
- Which real buyer questions would you test first, and who approves them?
- Which AI systems and modes are included, and what are the limitations of the data?
- How will you classify a citation versus a brand recommendation?
- Can I inspect a redacted example of the raw observation log and final report?
- What access to our website, analytics or proprietary data is genuinely needed?
- Which deliverables are one-time, and which require recurring work?
- Who will implement changes and verify them in production?
- Which parts of your proposal are observations, hypotheses or estimated opportunities?
- How do you measure progress without implying you control AI systems?
- What happens if the sampled answers do not improve?
What are the warning signs of a weak offer?
Be careful with claims of a permanent “number one ChatGPT ranking,” universal proprietary scores without transparent definitions, unsupported percentages, fake client screenshots, automated pages written only for keywords, unexplained backlink schemes and dashboards that combine mentions, citations and referrals into a single success count. Also be cautious if an agency refuses to provide a written scope, wants unnecessary account credentials or promises conversions it cannot attribute.
An agency that uses software is not inherently weaker than a manual agency. Automation can help with repeat capture and aggregation. The question is whether outputs are checked, the sampling is honest and recommendations follow from inspected evidence.
How should you compare prices and contracts?
Compare the scope, not only the monthly fee. A lower-price snapshot with ten prompts may not be equivalent to a human-reviewed investigation that checks dozens of sources and implementation issues. Ask whether quoted costs include prompt research, platform access, localization, competitor analysis, technical audits, editorial work, deployment, monitoring, client meetings and follow-up measurement.
Require clear ownership of the reports, spreadsheets, updated website files and analytics accounts. Define who approves claims and content; whether subcontractors are used; how confidential prompts and business data are handled; cancellation terms; and whether continued monitoring is optional. Avoid paying recurring fees before confirming what is delivered each cycle.
Should you hire an agency or build an in-house team?
An internal team can be suitable when you already have technical SEO, analytics, editorial and research capacity and need frequent coordination. An external specialist may provide useful independent testing, frameworks and implementation experience. A hybrid arrangement often works: the agency diagnoses, trains or assists with specialized analysis, while internal experts own product accuracy and publication approvals.
Choose based on available people and the quality of the work plan. Neither route gives direct control over third-party AI answers.
How can you measure whether the engagement worked?
Agree separate outcome layers before work starts. Delivery outcomes include verified fixes and accepted reports. Visibility observations include sampled mentions, recommendations and citations within stated denominators. Business outcomes include qualified visits, enquiries and revenue where attribution is defensible. A favorable source citation is not automatically a commercial conversion, and an improved answer cannot alone establish causation.
For the actual calculation method, see our AI visibility measurement guide. For independent checks before paying a provider, see how to check AI visibility. For a transparent research approach, read our manual audit methodology.
Frequently asked questions
Is an AEO agency different from a GEO agency?
The labels overlap, and agencies define them differently. Compare actual answer-design, source analysis, implementation and monitoring deliverables rather than titles alone.
Can an AI visibility agency guarantee ChatGPT recommendations?
No responsible provider can guarantee stable unpaid recommendations in an independent AI system.
Do I need a manual audit before ongoing monitoring?
A defined baseline is helpful before recurring monitoring, whether created internally or by an agency. It prevents confusing changing prompts with a real trend.
What proof should a new agency provide if it has no client case studies?
Look for genuine team qualifications, an honest methods document, an appropriately labeled demonstration, source-level evidence and achievable deliverables. A lack of public cases should not justify fabricated results.
Are AI visibility tools enough without an agency?
For some teams, yes. Tools can organize observations, but someone must interpret intent, inspect citations, validate website changes and understand the commercial context.
Should I pay based on mentions or leads?
Agree what each metric captures and what can be verified. Mentions can indicate sampled visibility, but lead outcomes depend on many factors and require trustworthy attribution.
How long should an AI visibility contract run?
Duration depends on the size of the audit, the implementation backlog and the reporting cadence. Ask for milestone-based deliverables and a clear renewal or cancellation policy rather than a universal timeline.
What should I bring to the first agency meeting?
Bring your main services, buyer personas, important markets, current website pages, representative sales questions, competitors and any available search or enquiry baseline.
Apply the checklist to us, too
Use these questions to evaluate MentionGauge alongside every other provider. We publish our research standards and explain our manual audit scope. Request a written quote that names deliverables and limitations rather than relying on an unverifiable AI ranking promise.
Know what AI answers actually show.
Request a scoped manual audit to understand how your business appears across relevant AI-assisted search experiences and what can be improved.