What should be in an RFP for an AEO agency?

What goals and scope should the RFP set?

Begin with the business decision behind the search. Do you want more qualified sales talks, clearer answers about your services, or better coverage in a market? Pick the goals your team can act on. Tell bidders which brands, sites, markets and customer groups are in scope. Also say what is out of scope. That lets each bidder build a plan for the same job.

Give each goal a plain definition. A qualified lead might be a person who fits your service area and asks for a service you sell. Agree on who makes that call and where it gets recorded. A mention in an AI answer is a different event from a visit, a lead or a sale. Ask bidders to keep those outcomes apart when they set targets.

Example: A business with sites in several cities wants more inquiries for a service. Its brief names the cities, the service pages and the sales team that will judge lead fit. It asks each bidder to state what it would measure and what data it needs.

Add the limits that shape the work. Tell bidders if a legal team must review claims, if local teams own their pages, or if a site change needs a separate release. Say which teams can provide subject experts and approve edits. Ask for a scope that fits those facts, with any gaps or assumptions called out.

Do not ask every bidder to promise the same result by a fixed date. Ask for a proposed review date, the basis for any forecast and the facts that could change it. Google says no one can promise a top rank. A useful bid shows how the agency will report progress and respond if the evidence is weak.

Which customer questions and AI tools should you include?

Give bidders a sample of the real questions buyers ask before they choose you. Draw them from sales calls, support notes and your own search data. Cover early questions, comparisons, price concerns and the final choice. Remove private customer details before you share the list. Ask bidders to flag gaps and explain how they would select the questions that matter most.

Name the AI tools and markets you want covered. A question about a national service can have a different intent from one about a local office. Ask bidders to keep the question text, place, date and tool with each observation. That record makes a later check easier to compare. It also helps your team see which questions matter to an actual buyer.

Example: A firm sells a complex service in more than one state. Its sample includes a broad choice question and a question about local eligibility. The firm asks each bidder to explain which page or fact should help a buyer with each need.

Ask for a baseline, meaning a record of what was seen before work starts. It should show the questions checked, the answers observed and any links to sources. A citation is when an AI answer links to a page or names a source. Note whether a check found a name, a link or both. Those are distinct observations.

An answer can vary with the wording and context of a question. A small set of checks is a sample, not a full count of what all buyers see. Ask how the agency will handle changes to the question set and preserve earlier results. A clear record is more useful than a single score with no way to inspect its parts. Google’s “Do you need an SEO?” guide recommends asking how an agency measures success.

How should the RFP define measurement?

Ask for a measurement plan that connects each goal to a named source of data. Start with the record before the work, then agree on when to review it. Define each measure, its owner and its limits. Request the raw data or a clear path to it, as well as a brief account of what changed. Your team should be able to check the result without relying on a black box.

Google’s generative AI performance report includes impressions from AI Overviews and AI Mode. An impression is an appearance counted by the report, not proof of a click or a sale. Google also says AI feature traffic is part of overall search traffic in its Search Console performance report. Ask bidders which report they mean and what each figure can support.

Bing offers an AI Performance report in public preview. It shows citation counts and sampled questions for covered experiences. Bing says those counts describe citation frequency, not rank or placement, and the report does not show clicks. Keep its figures separate from Google figures and from your own site visits. They do not measure the same event.

Example: A bidder reports that a page appeared in a platform report and that your site received an inquiry. Ask for the dates and sources behind both facts. Unless the lead record supports the link, the report should not claim the AI appearance caused the inquiry. Your sales team can still review whether lead quality changed.

Ask how a bidder will count an AI referral, how it will handle unknown sources and whether your privacy rules permit the proposed data use. A buyer may move across devices or return later without a visible trail. Your own CRM, the system that tracks leads and sales, can add context. It cannot fill every gap in the path from answer to sale.

If you use a score, ask for its formula and the events behind it. Ask what happens when a source is missing or a platform changes its report. A useful score should lead to a decision, such as checking a page or reviewing a claim. It should not replace the source record or suggest access to a platform’s private systems. Keep the source record for the Google generative AI performance report and Bing AI Performance report.

What work, owners and approvals should a bid name?

Ask for a list of deliverables tied to the goals in your brief. A deliverable is an item you can review, such as a page brief, a set of edits or a report. Each item should have an owner, an expected handoff and a way to accept it. Ask what your staff must provide and what happens when access or approval is delayed.

Name the people who can approve claims, site changes and public content. Give each bidder the same view of those limits. Ask who will do the work, who will lead the account and whether any part will be passed to another firm. The bid should name the role responsible for each handoff, including work that stays with your team.

Example: A firm has a central brand team and local service leads. Its RFP says local leads check service facts while the brand team approves public copy. A bid can then show where each review fits. That avoids an assumed approval path that no one on the client side has agreed to follow.

Ask for samples that show the work and the result, with enough context to judge fit. A case study should say what was measured, when, and what else may have affected the outcome. Ask whether the agency had permission to share the example. One success does not establish a typical result or prove that the same result will happen for you.

Ask how the agency will explain proposed site changes and keep a record of them. Google recommends asking for both the changes and the reasons behind the advice. Your team should know who can publish, who can undo a change and where the final copy lives. Set a route for disputes over facts, brand voice or risk before work begins.

Your brief can ask bidders to work with existing search and content teams. Google treats work on visibility in its generative AI search as part of search optimization. Avoid paying twice for the same item by asking each bidder to mark shared work and the owner of the final decision. A boundary on paper helps teams act together.

What contract terms and safeguards should you request?

Ask every bidder to use the same cost schedule. It should show the fee, what it covers, any setup or outside costs, and the terms for extra work. Cost drivers include the number of sites, the number of markets, who makes site edits, and who approves copy. Ask each bidder to show which of those items are in scope and which your team must handle. Compare the full scope before comparing the total.

Ask for the term, payment schedule, renewal terms, notice period and exit process. The bid should explain what happens to work in progress if either party ends the agreement. Identify what you own, what you may keep using and what remains subject to a license. Have your own counsel review the terms for your needs.

Example: A bidder proposes a fee for a set scope but leaves edits by your web team out. Another includes those edits. Ask both to show that difference in the same cost schedule. You can then judge the total work and the burden on your staff, rather than treating unlike bids as equal.

Cover access and data in the contract. Name the accounts and types of data the agency needs, the people who may use them and the person who grants access. Ask how access is removed at the end and how your data and working files are returned or deleted. Your security and legal teams can set the rules that fit your business.

Include a change process. If a new market or site is added, ask for a written scope and cost before that work starts. Set a path for missed handoffs, delayed reviews and a dispute over acceptance. The goal is a record both sides can use, not a vague promise to sort it out later.

Put guardrails around claims and search practices. Google warns against paid links meant to manipulate rankings and large volumes of pages made mainly to manipulate results. Ask bidders to disclose paid placements and any work that could put your brand at risk.

How should you compare bids and review the work?

Send the same brief, sample questions and answer format to each bidder. Ask for the same scope, assumptions, team roles, measures and contract terms. Allow space for a bidder to suggest a different route, but ask it to price and explain that option separately. This makes a fair comparison possible without losing useful ideas.

Use a scorecard before proposals arrive. Choose the factors that matter to your business, such as fit with your goals, quality of proof, clarity of measurement, team ownership and contract terms. Give each factor a weight your team accepts. Record the reason for each score and any unanswered question. The score should support a decision, not make it for you.

Example: A review team likes a bid with a polished report but cannot find who approves changes. It asks for the missing owner and checks whether that person has time. Another bid gives a clear handoff but lacks a source for a claimed result. The team asks for the source before scoring the proof.

Use the same discipline after you hire. Set a regular review in the agreement and bring the people who can make decisions. Read the work against the agreed scope, not just a slide with a rising line. Ask what was delivered, what the evidence shows, what remains unknown and what the team proposes to change.

Keep a decision log with the owner and next review date. If a result is weak, ask whether the goal, question set, page, measurement or approval path needs attention. If a change is proposed, record its cost and effect on the rest of the scope. Do not treat silence from an AI tool as proof of a penalty.

Watch for bids that promise a rank or claim secret access to a platform. Outside tools cannot see Google’s internal ranking or AI systems, according to Google. For more questions to use before signing, see the related guide below. Ask for a source and a plain explanation when a claim sounds certain. Score every bid against the seven-item list below before the decision date.

What should you check before sending the RFP?

• A business goal and a clear scope for each brand and market
• A sample of buyer questions with private details removed
• The AI tools and places included in the review
• A baseline and a definition for each result
• Named owners for the work, reviews and site access
• A cost schedule with extra work and exit terms
• A fair scorecard and a date for the decision