Choose a GEO agency that can show how it investigates customer questions, improves useful business information and checks the results. Then establish who will do that work for your company. A screenshot of an AI recommendation or an impressive visibility score cannot answer those questions on its own.
For a Dubai business, the strongest fit is a team that understands your real services and customers, can produce accurate content, and can get agreed improvements onto your website. Its explanations should remain clear when you ask about an unsuccessful result as well as a successful one.
Start with evidence you can inspect
Ask the agency to show a relevant piece of work and explain its own contribution. A result for a business with similar customer decisions can be more useful than a famous logo with no explanation.
Separate three kinds of evidence:
- Work delivered by the agency for a client.
- Work a team member completed in a previous role.
- A demonstration or sample created to explain the approach.
All can help you assess capability, but they prove different things. A demonstration does not establish a client result, and joining a project after a business was already appearing in answers does not establish that the agency created that visibility.
Respect confidentiality. A redacted example can still show the question, dates, work completed, source evidence and reasoning. You do not need access to another client's private analytics or customer records to judge whether the explanation is credible.
Ask to see both a successful and an unchanged result
Choose one claimed improvement and ask the team to walk through it:
- What customer decision did the question represent?
- What did the answer show before the work, and under what conditions?
- What did the agency actually change, and when did it go live?
- What was observed afterwards, using a comparable check?
- What else changed that might help explain the result?
Then ask for a question where the work produced no clear improvement. What did the team investigate? Did it change the plan, continue observing, or decide the question was not worth pursuing?
You are assessing judgement. A team that can discuss an unchanged result and a sensible next action gives you more to evaluate than a deck containing only successful screenshots.
A later mention is not, by itself, proof that one edit caused it. Equally, an unchanged answer does not automatically show that the work was pointless. Look for a reasoned explanation tied to what was delivered and what the evidence supports.
Ask exactly how the agency measures AI visibility
Ask for a sample report that makes these details clear:
- The actual questions, with branded and unbranded prompts identified separately.
- Platforms and experiences covered, separating Google AI Mode from AI Overviews where relevant.
- Language, location context, date and whether web/search mode was used where a platform offers different modes.
- How often the same or equivalent questions are sampled, and how failed or missing responses are handled.
- Changes to the question set, with a record of what changed and when.
A combined AI visibility score can be useful if you can understand its method and compare like with like.
Here is a deliberately simple hypothetical example. A business appears in 10 of 20 checked answers: 50%. If the next report keeps only the same 10 answers that mentioned it, the result becomes 100% without one additional mention. The percentage improved because the comparison changed.
This is not a claim about any particular agency. It shows why you need the underlying questions and counting rules, not just the headline chart. If the question set expands or changes, the report should explain that change rather than present the scores as directly comparable.
Ask the team to distinguish six things: a mention names you; a recommendation suggests choosing you; a citation attributes information to a source; a cited URL identifies the referenced page; a referral visit reaches your website; and a business outcome records what happened commercially. A neutral or critical mention is not a recommendation. None should be counted as a sale merely because it appears in a dashboard.
Ask what comes from the agency's monitoring and what comes from platform data. For example, Bing Webmaster Tools AI Performance reports citations, cited URLs and sampled grounding queries across supported Microsoft and partner experiences. It does not cover every AI platform or provide an AI ranking. Bing's explanation of AI Performance.
A proprietary dashboard can organise useful observations. It does not establish privileged access to a search engine. Google explicitly says third-party tools do not have access to its internal ranking data and cannot guarantee performance. Google's guidance on third-party tools and advice.
Check whether the team can diagnose discovery problems
Ask: “How would you check whether our important pages can actually be found?” A useful answer covers crawl access, indexability, the correct canonical page and relevant Google, Bing and AI search crawlers. The team should explain whether a problem needs a content change, a website fix or further investigation, and who will handle it.
For ChatGPT Search, OAI-SearchBot concerns search discovery; GPTBot concerns model training. Their controls are separate. Allowing search access does not guarantee citation. OpenAI's crawler documentation.
Treat claims that Google AI requires llms.txt or special AI schema as a red flag. Google requires neither; llms.txt does not affect its Search visibility or rankings. Google's current guidance. A file maintained for another system needs a separate explanation of its purpose.
You do not need to test the salesperson's vocabulary. Ask the delivery team to explain what its checks mean for your website in plain English.
Meet the people responsible for the work
Find out who will select questions, verify company facts, edit content, handle technical issues and explain results. These responsibilities may involve several people; the important point is that they are covered and coordinated.
Ask who reviews AI-assisted drafts. Can that person recognise an incorrect service claim, distinguish an opinion from a sourced fact, and obtain the right approval before publication?
Then ask what happens during an ordinary difficult week: your reviewer is unavailable, a page cannot be published, or a report finds a material error about your business. Who owns the next action, and how will you know it is moving?
The person selling the programme may not be the person running it. Meet the delivery lead and understand how any specialist or subcontracted work is supervised.
Test whether the Dubai brief changes the plan
Ask the agency to describe your actual market, not simply add “Dubai” to a global question list.
For a product business, useful questions might depend on UAE availability, the currency and conditions of the offer, local delivery coverage or whether buyers can visit a real showroom. For a service business, actual coverage and service limitations can determine whether a recommendation is suitable.
If your customers research in Arabic as well as English, ask who will research and review each language. A translated English question set is not evidence that the team understands the questions those customers ask. Confirm the required capability and scope instead of assuming it from a bilingual sales presentation.
Local office details alone do not prove this understanding. Look for a team that asks informed questions about your customers and checks business facts with you.
Ask how the agency earns attention
Useful content should contribute something a buyer can use: accurate service details, a clear comparison, an explanation based on experience, or authorised evidence of work. Ask what the proposed content adds beyond pages already available online.
AI assistance is not the issue by itself. Google's scaled-content policy targets large amounts of low-value content created primarily to manipulate rankings, regardless of how it was produced. Google's spam policies.
Ask equally clear questions about external activity. Does it mean correcting legitimate profiles or directories, editorial outreach, digital PR, paid promotion, or participation in relevant communities? Which activity is proposed, why is the source relevant, and who approves it?
Reject invented endorsements, fake reviews, spammy community personas and undisclosed promotional identities. A paid article should never be presented as a guaranteed AI recommendation. You should be comfortable with the methods used in your company's name.
Keep accountability and access clear
Before choosing, establish who approves content, who can make website changes and how completed work is recorded. Keep company control of the domain, website and relevant business accounts, with appropriate working access for the agency.
Ask what you receive if the engagement ends: commissioned and published content, reports, the agreed question set, historical observations, source and citation records, and the change history. Ask what can be exported and in what usable format. Clarify rights and handover arrangements before relying on an agency-owned dashboard as your only record.
Price matters, but compare candidates on the same required work. A team that provides advice alone and a team that also publishes improvements are different delivery arrangements. Decide which your business can actually support.
When Lunasol is a strong fit
Lunasol is worth considering when you want AI visibility work connected to content delivery and practical website improvements. Its GEO offer brings together buyer-question research, accuracy checks, company-information consistency and new or improved content.
Research, human editing, factual review and CMS publication can be included in the agreed content scope. Related SEO and website work can also be coordinated when commissioned. This is particularly useful if your business does not have an internal team ready to turn findings into published changes.
Apply the same evidence standard to Lunasol: ask how the proposed questions relate to your buyers, what work will become live, who will review it and how the findings will shape the next actions. The attraction should be a clear delivery plan and accountable work, rather than a promise to control an AI system.
Discuss your business with Lunasol's GEO team. Bring a priority service, your website and the customer questions you want the programme to address. That gives both sides a useful basis for deciding whether the fit is right.