How do you run a 90-day AEO pilot before a full contract?

Which buyer questions belong in the pilot?

A useful pilot starts with the choices your buyers face. Pick a small set of real questions that fit the service, market and buyer you want to reach. Answer engine optimization, or AEO, means making your business easy for AI tools to find, trust and name in their answers. A broad question set may look busy while hiding the choices that matter to your business.

Ask sales and service teams what buyers ask before they make contact. Include questions about fit, risks and alternatives when they match your offer. Write each question as a buyer would ask it. Keep the exact wording, the market and the AI tool beside it. Agree on the list before work starts, so a later report does not swap hard questions for easy ones.

For a service in one city, include the service and place in the question, then add a question about how a buyer should judge fit. Keep the wording in the plan so the same question can be used at both checks.

Choose who will own the list on your side. Ask the agency to flag unclear terms before the start check. If a question has several meanings, agree on the intended one. You can keep a separate list of ideas for later without changing the pilot set. That makes the end review fair to both sides.

Do not make the list so narrow that one favorable answer decides the whole deal. Do not try to cover every product and region either. Pick a scope your team can review and approve. Write down what is out of scope, especially when other brands, sites or markets could change during the same period. Walk away if the agency will not show you the question list. The finished plan should name the owner, the agreed questions and what is out of scope.

Ask each person who will use the result to review the list before the start. A sales lead may spot a question that sounds natural but does not match a real buying decision. A service lead may see an unclear term. Resolve those points in the written list, then save a dated copy for the end review. If someone proposes a new question later, keep it separate so the original comparison remains easy to follow.

How do you set a useful start check?

A start check is the record you will use for comparison at the end. For each agreed question, save the exact prompt, tool, date, market and answer. Note whether your brand is named and whether a source is linked. A citation is a link to a source, or a name given as the source of an answer. Save the source and the surrounding answer, not just a count.

Use the same setup at the end as far as the tool allows. Record the account, location and any settings that could affect the answer. If a tool or setting changes, note the change beside the result. AI answers can vary, so a single saved answer is an observation, not a full view of all buyers.

On day 1, save the answer to each agreed question. Keep the full answer and its source alongside the date, tool and market. On day 90, repeat the check and put the records side by side.

Also record the site and business measures you plan to review. These might include visits to relevant pages, inquiries and the quality of those inquiries. Agree on the date range and the source for each measure. Record gaps in access before the pilot begins, rather than filling them with guesses at the end.

Give the buyer and the agency access to the same record. Include a note field for a changed page, a site outage or a new offer. Record the day 1 check and the planned day 90 check in the shared file.

Before the check, agree on who saves the answers and where the files will live. Check that the people at the end meeting can open them. If an answer names your business but describes the wrong service, save that wording too. If it names another source, keep the source with the answer. The record should make it possible to read the result without relying on a chart alone.

What should the agency deliver during the pilot?

Ask for a short written scope with named outputs, owners and due dates. The work might cover a review of existing pages, proposed changes, approved content or fixes to facts about your business. Choose the work that fits the questions in the pilot. A deliverable is something you can inspect and accept, such as a page ready for review or a list of proposed changes.

State who can approve claims, who can publish changes and who supplies proof. Ask what happens if a page needs review by legal or a product team. Set a path for missed dates and open questions. An agency cannot deliver a live change if the person who must approve it never sees the draft.

Name the person who checks product facts and the person who can approve a page for publication. A draft waiting for review is a different output from an approved page that is live.

Ask for a regular report that shows completed work, work awaiting approval and the next decision. It should link each output to the buyer questions it is meant to help address. Ask the agency to explain any change in scope before doing extra work. Make room to record delays on either side without turning the report into a sales pitch.

Google suggests asking a provider about past work, expected results, how success is measured and how you will communicate. Those are useful questions for an AEO pilot too. Ask for an example report and walk through it with the people who will use it. A clear report should let them tell what was done, what was observed and what remains unknown. Put a named owner and due date beside every action.

Do not sign without a named owner for each action. Separate the agency’s work from the work your own team must do. If your team must provide source material, name the person responsible and the date it is due. If a page needs a fact check, state who will make the call. At the review, use the scope to mark each output as delivered, awaiting approval or still open. Ask for a written explanation of any missed item.

How should you read pilot results?

Look at the saved answers first. For each question, compare the start and end records. Note whether the brand was named, whether the answer described it correctly and whether a source was cited. Read the full answer. A mention in a poor fit or an incorrect claim may call for a different response than a useful citation.

For the details on interpreting platform reports, see the AEO agency monthly-report guide.

Keep site and inquiry data as separate evidence. A change in inquiries alongside a saved AI answer does not, by itself, prove the answer caused that change. Review inquiry records for context and identify the source and date range for each measure.

Put the start and end answers side by side for each question. Mark a correct mention, an incorrect description and a cited source separately. Read the surrounding answer before deciding whether a mention is useful. If a source is missing from the record, ask for it before treating the observation as complete.

Review delivery as well as visibility. List the pages or other outputs accepted, those still waiting for approval and any work that went live late. Match each output to the questions it was meant to address. This helps the group discuss the work that actually took place before deciding what the next scope should cover.

Keep raw observations apart from your judgment. Record missing data and changed settings next to the affected comparison. Bring the dated answers, accepted work and open items to the end meeting.

A useful review can still surface an issue when the answer names your business. Look at whether it describes your service and market accurately, and whether the linked source supports that description. Give an incorrect claim its own action and owner. If the record is incomplete, ask for the missing answer or source instead of guessing. Save the final comparison with the decision record.

What decision rule should you write before work starts?

Write the rule before the first check, while neither side knows the end result. Define what would support a longer contract, what would call for more proof and what would lead you to stop. Use both work you can inspect and the direction of the measures you agreed to review.

For a continue decision, you might require agreed work to be delivered and accepted, clear reporting, and enough useful movement or learning to justify the next scope. Define what useful means for your own business before you begin. Avoid a rule based on a single answer or a sales total that the pilot cannot trace.

For a revise decision, name the gaps you would need to close. Perhaps work went live late because a review took longer than planned, or a data source was not available. Decide who will supply the missing information and when you will review it. Do not treat an open question as a confirmed win.

Put the continue, revise and stop choices in the same document as the scope. Under each choice, list the evidence the decision makers will review and the person who can make the call.

For a stop decision, consider missed commitments, unclear ownership or a report that cannot support its claims. Ask for the final files, access details and any open work. Record what you learned even if you do not continue. You can use that record when you compare a future proposal or decide what your own team can do.

Google says no one can guarantee a first-place Google ranking. Apply the same care to promises about AI answers: ask what can be delivered and checked, rather than accepting a promise that a third-party system will name you. Confidence is useful when it comes with a clear scope and honest limits. Bring the written continue, revise and stop rule to the day 90 review.

For each path, decide what happens after the meeting. A continue decision needs a proposed next scope and the person who can approve it. A revise decision needs a list of missing evidence, an owner and a review date. A stop decision needs a handoff of the files and open work. Write these actions beside the decision, so the team does not confuse interest in a longer deal with approval of one.

What should be in writing before the pilot starts?

Put the question list, start check, scope, review dates and decision rule in one place. Name a lead on each side and the people who can approve work. State the agreed scope and terms in your agreement. Ask how added work is approved and how either side can end the pilot.

Also address who can use the content and records after the work ends. List the accounts involved, who controls access and how access will be removed. If your agency needs business or customer data, define what it needs and who may see it. Have your own counsel review the terms for your situation.

If your web team needs to keep working with approved pages after the pilot, put the proposed use and account access in the agreement before signing.

Plan the end meeting before the start. Invite the person who owns the budget, the person who approves content and the person who can explain the data. Ask for the report in time to read it before the meeting. The goal is a decision on the next scope, a request for missing evidence or a clean stop.

Keep a record of changes to the plan. If a question, tool or due date must change, note what changed and why. Preserve the original plan alongside the update. At the end, compare like with like where possible and mark the places where you could not. That is more useful than a neat chart that hides a change in the test.

A pilot also helps you judge the working relationship. Did the agency explain its claims? Did your team have time to review work? Were issues raised early enough to address? These are buyer judgments, not platform metrics. Record the reasons for the decision so the next owner can understand them. Save the decision, its reasons and the next owner in a dated record.

Before the work begins, send the plan to the people who must approve it. Check that the dates for content review and the end meeting work for them. Ask the agency to identify any access it needs and record who controls each account. When the plan changes, keep the earlier version as well as the update. At the end meeting, assign an owner and due date to each remaining item and share the dated record with the decision makers.

What should you check before signing?

• Buyer questions and target market are agreed
• Start and end checks use the same saved questions
• Work, owners and review dates are written down
• Reports distinguish observations from business outcomes
• Decision rule covers continuing, revising and stopping
• Content use, account access and exit terms are clear