Updated on 16 November 2026.
Pilot a GEO service with your first client
Before offering AI monitoring, pilot it with one client whose offer and typical sales process you can describe. Agree on the decision the project should support. For example, a retailer wants to know whether AI recommends its products in comparisons across two categories, and the team has time to improve one page in the next cycle.
Include measurement, data checks, and agreed tasks. Record who reviews the answers, who verifies product information, and who updates the website. When comparing tools, ask the analyst who will run the project to produce one record with a full answer and its cited source. Time the work, because the monthly cost includes the analyst’s time.
For one client, start with 12 buying questions in one language, two AI systems, three competitors relevant to the categories, and one client review. Treat the numbers as a starting point. For two categories, use separate buying criteria so recommendations for different purchases are not averaged together.
| Stage | What the client receives | Acceptance criterion |
|---|---|---|
| Scope | A record of the market, offer, and questions | Each question reflects a buyer need or decision. |
| First measurement | Answers and sources in context | Examples can be reproduced using the date, language, and system name. |
| Review | One task based on a verifiable fact | The task names the URL, owner, deadline, and acceptance criterion. |
| Decision on continuing | Recommended service scope and labour cost | The team can complete a full cycle in the time allowed. |
- Agree with the client on what you will implement during the pilot.
- Check how the plan separates clients, which answers are archived, and what measurement limits apply.
- Before closing the pilot, record what you will measure again and who will approve the next scope.
AI visibility audit before sending a proposal
How to budget for AI monitoring
Start with a client record
One workflow can organize many projects, but each client needs its own scope. Record the brand, domains and aliases, market, language, offering, project owner and analysis objective. Separate brand-awareness questions from supplier-selection questions. Centralization means shared working rules; it does not mean combining different clients’ answers.
Share intent templates, maintain separate prompt sets
Your team can share question templates, then adapt them to each client’s actual offering. “Which supplier installs furniture in Kraków?” tests a different issue from “Does brand X offer installation?”. Group questions by problem, comparison, purchase and constraints. Give every set a version, date and reason for changes. Keep a stable core for comparisons over time.
Define competitors and comparison conditions
The client’s competitor list may differ from the brands AI mentions. Keep both lists and record their origin. Compare brands within the same answers and identical scope. Record the AI system, country, language, date and available settings. Do not compare a client measured using generic questions with a competitor tested only through questions containing its name.
Turn an answer into a task
A mention, recommendation and source citation are separate events. Open the answer and check the cited URL before choosing an action. Every task needs evidence, an existing URL, an owner, a deadline and an acceptance condition. Brand absence may reflect a content gap, an accessibility problem or a poorly matched question; the score alone cannot identify the cause.
Did updating your article improve AI visibility? How to plan a before-and-after test
Example: three projects, one standard
Illustrative example: a shop needs clearer delivery terms, a service business needs coverage information, and a SaaS provider needs an integration description. The agency uses the same task record with different sources and owners. Acceptance means approving accurate information and a working URL. Any subsequent visibility change needs separate measurement; completing a task does not demonstrate an AI effect.
One standard, separate client actions. Example organisation of three agency projects. Illustrative example. Each action: owner + deadline + acceptance.
Can another account manager take over the client’s monitoring?
Use one buyer decision as a handover exercise. Imagine two clients who both sell coffee equipment. One rents machines to offices with maintenance included. The other sells refurbished machines outright. The account card should tell a new reviewer which offers qualify, what market is in scope and which public page confirms the service terms.
Give the incoming reviewer a saved answer without its score. Ask them to identify the supplier recommended, the buying condition behind that choice and the source supporting it. The account owner checks the interpretation. A service specialist resolves open questions about maintenance before an editor changes any copy.
Accept the handover when the reviewer can reproduce the score, find the relevant page and assign the next action to the right person. Keep shared scoring rules in the agency template. Store each client’s prompts, evidence and delivery notes with that client’s project.
Use Semly within the agency workflow
Semly Agency supports client projects in one account, with shared pools of monitored questions and articles. Resource limits can be assigned to individual projects. The public offer also describes team and client access management and white-label reporting options. Confirm the scope of the service you choose. Resource allocation, user permissions and content approval are separate decisions.
Allocate prompt and article resources across client projects in Semly Agency
Blog and knowledge base without cannibalisation: how to divide topics for SEO and AI
Assign responsibility and access boundaries
Name the people responsible for project configuration, answer analysis, change approval and client communication. Separate client data in filenames and exports; check the correct brand before sharing a report. A common standard does not require identical access for everyone. If the tool lacks the necessary permission separation, arrange it in the process and storage location. Record who may change prompts and approve scope changes in the project profile.
Introduce a review cycle and change log
Before each analysis, check measurement completeness and changes to the client's offering. Then review recurring errors, new sources and competitor differences. Select a manageable number of evidence-backed actions with owners. Log prompt, competitor or system changes with dates and reasons; label old and new series separately. Regularity supports comparability, while frequency should match change speed and team capacity. Do not increase question counts at the expense of quality control.
Agency report template: one record per client
Copy this record into your document or spreadsheet. It is a blank template, not a Semly account report or an automatic connector. Remove other clients' information and internal comments before sharing.
Record scope before numbers: project, period, country, language, exact AI system and question-set version. Define each unit and denominator. Count answers containing the brand rather than repeated mentions. Failed collections are not valid answers without the brand.
| Item | Record | Acceptance condition |
|---|---|---|
| Scope | Project, period, country, language, system, question version | Matches the source view |
| Mention and citation | Separate answer counts and a common denominator | Different events stay separate |
| Action | Existing URL, owner, deadline, criterion | Approved next step |
| Sharing | Recipients and permitted scope | No other clients' information |
Brand-answer share = answers containing the brand / valid answers. Calculate own-URL answer share separately on the same sample. With no valid answers, report missing data rather than zero. Do not average clients' percentages or incompatible samples without agreed definitions.
CSV column names stay fixed for moving between spreadsheets. Each row represents one consistent scope. Enter approved information only; an empty cell means no entry, not a zero result.
project,period_start,period_end,country,language,ai_system,question_set_version,valid_answers,answers_with_brand,answers_with_own_url,scope_notes,action_url,owner,due_date,acceptance_criterion,approval_status
,,,,,,,,,,,,,,,Accept the report and its integration
If your plan includes a reporting integration, select the correct project and period, then compare scope and definitions with the source view. Check available fields, refresh frequency, market and system filters, and isolation from other projects. Publishing this template does not activate a Data Studio connection.
Deliver a decision: what to change, on which existing URL, who owns it and what closes the task. Client acceptance covers the report and recommendations. It does not establish increased visibility or sales.
Choose the reporting route before implementation
A report template, an export file and a Data Studio connection are three different routes. Use the one actually available in your configuration. A manually completed CSV does not establish that a connector or automatic refresh works.
For a manual report, record the source, scope and preparation date, and assign responsibility for updates.
For a reporting connection, confirm authorization, fields, types, filters and refresh rules. Use the connection documentation for names and aggregations.
Before sharing, inspect the report with the recipient's permissions and confirm that it shows only the intended project. Do not grant broad access simply to make the report work.
Send missing specifications to the integration owner. Do not replace them with a guessed export schema. The templates in this article organize the work; they do not configure a connection.
What does Google AI mean in this pricing table?
The Google AI entry covers Google AI Overviews and Google AI Mode. Gemini is a separate entry. Availability follows the systems listed for each plan. Do not treat Google's products as a single model or infer access to systems the plan does not list.
Check your company’s AI visibility
See which questions mention your brand and where to start investigating.
FAQ
Should every client use the same prompts?
You can share a framework for organising intent. Adapt the questions to each client’s offer, market, and buyers’ decisions.
Can you compare clients’ average results?
Compare them when definitions and measurement scope match. For different industries, markets, or question sets, reviewing projects individually is usually more useful.
When should you expand monitoring?
Expand when the team uses its existing results and a new market, category, or model meets a specific client need.
What should you show the client at the end of the pilot?
Show a full answer to their buying question, a verified source, and a specific change with an owner. Include the sample and the date of the next measurement so the client knows what the ongoing service will deliver.



