Your competitor appears in an AI answer. You don't. By the following afternoon, someone is offering to fix your “entity authority”.
Excellent. You had one problem. Now you also need a translator.
GEO, or generative engine optimisation, aims to improve how your business appears in AI-generated answers. It can be useful work. The buying mistake is paying for an impressive appearance without understanding what changed, whose question it answers or whether a potential customer can act on it.
A good proposal should connect a buying question to specific work and a measurable result. Here is how to put one through its paces before the invoice arrives.
Start with a customer you would actually want
Imagine a hotel selling meeting packages in Westlands.
“Best hotels in Kenya” might produce an exciting screenshot. It tells the events manager little about reaching someone organising a staff workshop for thirty people.
Ask an agency to bring ten initial buying questions, not ten thousand generated prompts. Beside each, request the likely buyer, their decision and the page that would help them proceed.
| Question to investigate | Decision behind it | Information the hotel needs to provide |
|---|---|---|
| Which Westlands venues can host a thirty-person training day? | Create a shortlist. | Actual room capacity, layouts and location. |
| Does the meeting package include lunch and equipment? | Compare the total package. | Inclusions, exclusions and current terms. |
| Can we book a workshop venue for this date? | Check availability. | A workable enquiry route and a person who replies. |
These are sample questions to validate against actual enquiries. The point is to make the agency explain why an appearance would matter to your business.
Keep branded and unbranded questions separate. “Is this hotel suitable?” tests a business already under consideration. “Which venue should I consider?” tests discovery. Combining them into one score can hide which job improved.
Make them show the starting point
Ask for the answers behind the score. A baseline should record the exact question, platform, date, available model or search settings and relevant language or location conditions. It should distinguish a mention from a recommendation and a source link.
For each result, check whether the business description is correct. Being recommended for a service you do not offer creates enquiries your team cannot use.
Use a repeatable core set of questions. For a small initial exercise, you might agree ten questions with three fresh runs each, producing thirty observations per platform. That is an example of a manageable test, not a statistically representative sample or an industry standard.
If you add questions later, report them separately. Otherwise, the score can improve because someone changed the exam.
Request a record of missing results too. A folder containing only favourable screenshots is a portfolio, not a baseline.
Turn the audit into a work order
Ask the agency to complete this sentence for every priority:
We found [specific problem] on [page or system]. We will change [specific item], and [named role] will verify completion by [date].
For the hotel, that could mean replacing a vague meetings page with accurate capacity, package details, real room photographs, and an availability enquiry. Another business may need inconsistent business information corrected or a technical access problem resolved.
The proposal should separate implementation from recommendations. If writing is included but uploading is not, someone still needs to publish the page. If the agency recommends a developer, establish whose budget pays for that work.
There is also a useful nonsense detector.Google's AI search guidance says existing SEO fundamentals apply to AI Overviews and AI Mode. There are no extra technical requirements, special AI files, or special schema needed for inclusion. Supporting pages must be indexed and eligible for a search snippet; eligibility does not guarantee appearance.
That applies to Google's features. For ChatGPT search, OpenAI's crawler documentation distinguishes search access through OAI-SearchBot from training access through GPTBot. Ask the agency what access change it recommends and why. “Let every bot in” is not a diagnosis.
A provider can propose experiments beyond documented fundamentals. It should identify them as experiments and specify what would justify continuing.
Follow the recommendation through to the enquiry
Open the page an interested visitor would reach. Can a buyer judge fit and take the next step?
For the hotel, check capacity, location, package contents, and how to ask about a date. Submit an internal test enquiry and verify that the right person receives it. If the page promises a reply, check who owns that promise.
Give three stages separate names:
- A contact click: someone selected the button.
- A received enquiry: the business actually got a request.
- A qualified enquiry: the request fits the service and can be progressed.
An agency should not report all three as the same achievement. Decide which it will measure and what your team must record.
Bring Attention Lasso a proposal and the offer you want discovered. We can help identify whether the proposed work addresses a real visibility gap and whether the destination is ready for the attention.
Read the small numbers behind the big percentage
Consider this fictional report, using comparable questions and testing conditions:
| Measure | Baseline | Later review |
|---|---|---|
| Recommendations in monitored runs | 6 of 60 | 12 of 60 |
| Identifiable AI-referral sessions | 18 | 31 |
| Received forms attributed to those sessions | 1 | 2 |
| Qualified enquiries from those forms | 0 | 1 |
The recommendation rate moved from 10% to 20%. That is a 100% relative increase, which sounds rather grand until you meet the denominator.
The report shows a more promising monitored result and one qualified opportunity. It does not measure every customer's AI experience or prove that the agency caused the change. Other campaigns and platform changes may contribute.
Ask which buying questions improved, what work preceded the change, and what decision follows. “Continue monitoring” is sometimes appropriate. It should come with a reason and a review date.
Keep AI visibility, identifiable visits and sales outcomes separate. Some people discover a business in an answer and return another way. Google also includes AI-feature traffic within its overall Search Console Web reporting; an agency should explain how it derived any claimed AI-only figure rather than labelling all organic growth “GEO”.
Check the experience behind the pitch
Request a comparable example showing the starting problem, work delivered, and outcome. A US software case can demonstrate skill without establishing what the same team will achieve for a Kenyan venue.
Ask a reference three practical questions: what did the provider deliver, what did your team still have to do, and what happened when results disappointed?
Google's agency hiring advice recommends checking references, discussing measurement and assessing market experience. Its warning about guaranteed Google rankings is also a useful prompt to scrutinise guaranteed organic AI recommendations.
Compare the full scope: research, writing, technical changes, implementation, monitoring and review. Identify tool charges and the work your team supplies. Agree on ownership, cancellation and the reports or configuration you receive at handover.
Send every shortlisted agency the same brief
You can adapt this for your next meeting:
We want more suitable enquiries for [offer] from [buyer and market]. Please propose ten priority buying questions and explain their relevance. Show how you would establish our starting position, name the first changes and specify who implements them. Include a sample report, full costs and handover terms. Separate documented practices from experiments and delivery deadlines from hoped-for visibility gains.
Compare the clarity of the answers before comparing the size of the retainers. A focused audit or a few page fixes may be a better first purchase than an ongoing contract.
Frequently asked questions
Is GEO different from SEO?
It changes what you examine, while sharing substantial groundwork with SEO. GEO asks how your business appears in generated answers, including recommendations and factual accuracy. Google's AI search features still rely on established SEO fundamentals. Ask for the additional work behind an additional fee.
Can an agency guarantee ChatGPT will recommend my business?
No agency can reliably promise a permanent organic recommendation across changing questions and users. It can commit to research, implementation, and transparent reporting. A guaranteed screenshot is much easier to deliver than a dependable source of customers.
How long should I give a GEO campaign?
Set delivery milestones first, then agree when to review observable effects. A repaired page can be delivered on a deadline; crawling, inclusion, and customer response are less predictable. Decide in advance what evidence would justify extending the engagement.
What should a monthly GEO report contain?
It should show completed work, comparable visibility checks, accurate descriptions, identifiable visits and useful enquiry outcomes where measurable. Include counts and denominators. Finish with the next action and its owner, rather than a visibility score left to explain itself.
Do I need special AI schema or an llms.txt file?
Not to qualify for Google's AI Overviews or AI Mode. Google specifies no special schema or AI text-file requirement. For other systems, ask for platform-specific evidence and the exact problem a proposed file would solve before paying for it.
Can I assess my AI visibility without buying a tool?
Yes. Test a small set of relevant buying questions, save the full answers, and record the conditions. Repeat consistently and across platforms. It is a limited monitoring exercise, but it can reveal inaccurate descriptions and missing information before you purchase broader reporting.
Buy work you can inspect
The best pitch should leave you knowing what needs fixing and how you will judge progress. Bring your website and the product or service you want more enquiries for. Attention Lasso can help turn that into a scope you can evaluate.
Keep pulling the thread
Related reading
Lovable wants to do your marketing. Should you break up with your agency?
Read next
Your dashboard is busy. Is your business?
Read next