← Back to blog
Mara Ellison

How to Choose an AEO Agency: A Hiring Checklist

Learn how to choose an AEO agency, check its measurement methods, compare proposals and demand reporting that connects AI visibility to business results.

AEOStrategy
Summarize withChatGPTClaudePerplexityGrokMarkdown
On this page

Choose an AEO agency by checking its measurement method before reviewing its content plan. Ask for a baseline across named engines, repeated checks of the same buyer prompts, and reports with saved answers. Then compare who will fix your pages, improve third-party sources and correct false claims about your brand.

Start with three candidates from our agency shortlist by specialty. Send each the same brief, competitor names and sample prompts. Ask each to return a proposed scope and an anonymized report. You can compare their judgment before paying for a full audit.

What should an AEO engagement include?

Agree on the purpose first. AEO work should help buyers find accurate information about your brand inside AI answers. Our guide to answer engine optimization covers the foundation. For hiring, translate that purpose into five parts of the contract.

Prompt research. Ask the agency to build questions from sales calls, customer support and search data. Separate discovery prompts from comparisons and branded fact checks. A question that names your company tests a different problem from a category question that leaves it out.

Require an approval step for the prompt set. Otherwise, the agency could fill the report with questions that mention your brand and make visibility look strong before any work starts.

Measurement across engines. Specify each product, market and language. Ask for a baseline before publication or outreach begins. Keep the original prompts available so you can compare later results against the starting point.

On-site fixes. Require a ranked list of pages to change, with an owner and acceptance criteria. Examples include correcting a pricing page, making a product comparison easier to understand, or resolving a crawler access problem. State whether the agency implements changes or sends instructions to your developers.

Third-party source work. Ask which outside pages the agency would investigate when an answer cites someone else. The scope could include correcting a directory entry or supplying evidence to an industry publication. Require the target URL and proposed action. Separate paid placements from editorial outreach, and retain approval over public claims.

Fact corrections. Give the agency approved facts about pricing, features and availability. Require a log of false claims, the answers containing them and the corrective work. Ask the team to check those claims again after updating the source material. Publishing a correction and finding a corrected answer are separate milestones.

Each part needs a named person on the agency side. Meet the analyst and delivery lead who will handle your account. Ask who covers the work during an absence and which tasks require your team's time.

How can you check an agency's AEO capability?

Use the proposal call to examine one measurement example. Ask the analyst to trace a headline number back to its prompts and saved answers. You should be able to follow the calculation without buying the agency's software.

Request a baseline across several engines

Ask for results split across the engines relevant to your buyers. A proposal might cover ChatGPT, Perplexity and Google AI Overviews. Another might add Gemini or Copilot. Require a reason for the choice and a price for any coverage you need to add.

Treat Google's products as separate surfaces. Ask whether "Google AI" means AI Overviews, AI Mode, Gemini or some combination. Get a measurement schedule for each. A total engine count leaves too much room for misunderstanding.

The baseline should name your competitors and separate branded prompts from category prompts. Ask the agency to explain the prompts it chose to leave out.

Ask how the team handles changing answers

Request the method for repeating prompts and combining results over a reporting period. Ask how the analyst records an answer that names you once and omits you on the next check.

Our explanation of changes between repeated answers gives you context for that conversation. During the interview, look for a method you could reproduce: fixed wording, recorded dates and a clear treatment of missing answers.

Ask whether the agency measures through a consumer interface or an API, and whether search is enabled. Have it document location, language and account context where those apply. Compare results collected under the same conditions.

Check how the agency chooses the next task

Give the candidate a scenario: your brand appears in category answers, but the answers cite an outdated third-party comparison and repeat an old price.

Ask what the team would investigate first, who would do the correction work and what evidence would show progress. A useful response connects the observed answer to a source and a task. Ask for the same level of detail in an anonymized client example, without requesting confidential information.

SEO experience still matters. The extra evidence you need is the agency's ability to connect that work to the answers you are paying it to improve.

What should AEO reporting include?

Attach this checklist to the brief and contract. Ask candidates to mark each row as included, optional or unavailable. Request a sample using their proposed format.

Reporting requirementEvidence to demandAcceptance check
Per-engine visibilityBrand mention rate for each named engine, split by prompt groupThe report states what counts as a mention and which answers enter the rate
Share of voice against competitorsYour brand and a fixed list of named competitors, with the calculationYou can reproduce the figure from the export
Citations and answer evidenceFull saved answers, cited URLs, prompts, engines and collection datesYou can open the evidence behind a reported win or loss
Trend across several runsRepeated checks of unchanged prompts across comparable periodsThe analyst shows a trend alongside individual answer examples
AI-referred traffic and conversionsReferral sessions and agreed conversion events beside visibilityThe report distinguishes measured visits from estimated business impact
Delivery cadenceDates for updates, the monthly report and the review callThe contract names the person responsible for delivery
Work completed and next actionsChanged URLs, publication dates, correction status and ownersYou can connect the invoice to completed work and the next priorities

Keep mentions and citations separate. An answer can name your brand while linking to another site. Ask for both your own cited pages and outside pages that discuss you.

Require the agency to state how it calculates share of voice. Specify whether it counts a brand once per answer or counts repeated mentions. Freeze the competitor list for comparisons, and annotate any change.

Traffic belongs beside visibility because you need to know whether visitors arrived and converted. Ask which referrers the analyst includes, how conversions are defined and where tracking has gaps. Do not accept a revenue estimate without its assumptions.

Agree a reporting cadence before signing. For example, request weekly delivery updates and a monthly performance review. Label that as your requirement. Ask the agency to explain whether its measurement frequency supports the comparisons it proposes.

An agency can use tools such as Citlyze for client reporting. Ask for read-only access to the underlying workspace where available, or dated exports you can retain. You should have access to your evidence after the contract ends.

How should you compare agency quotes?

Most agencies do not publish prices. Use our breakdown of AEO budgets for published examples and numbers, then request quotes against the same scope. Treat those examples as context when you review a proposal.

Ask each candidate to separate the setup fee from recurring work. Itemize measurement, page updates, new content and outreach. Confirm which software subscriptions sit inside the retainer and which you must buy.

Compare capacity in concrete units. "Content optimization" could mean reviewing a page or rewriting and publishing it. Specify the number of pages, review rounds and implementation responsibilities. For outreach, request research and contact deliverables. The agency cannot set an editor's acceptance decision.

Ask what happens during a month with no visibility improvement. Require a written account of work completed, what the analyst learned and what the team plans to change. Agree how you will judge the engagement if the numbers stay flat.

Check the minimum term, notice period and handover terms. Name who owns the content, prompt library and measurement history. Ask how much time your subject experts and developers must reserve. Include that time when comparing the total cost.

Which promises should make you walk away?

Guaranteed rankings in ChatGPT. Reject a contract that promises a fixed organic position or guaranteed mention rate. Ask for commitments to coverage, completed work and evidence. Decide on renewal using the method and results you agreed to review.

Single-screenshot reports. A screenshot can illustrate an answer. Demand the prompt set and repeated checks behind any performance claim. Ask the agency to show an omission alongside a successful answer, then explain how both affect the reported figure.

No named engines. Require product names and schedules in writing. If the proposal says "AI search coverage" and the team cannot explain what it measures, you cannot check whether the service matches your brief.

"AI schema" upsells. Ask the agency to identify the markup, explain its purpose and cite the engine's documentation. For AI Overviews and AI Mode, Google's AI features guidance states: "There's also no special schema.org structured data that you need to add."

Require a specific explanation for any markup work, such as correcting structured data that conflicts with visible page content. A package called "AI schema" gives you no way to assess the task.

Google also retired FAQ rich results on May 7, 2026. Reject a proposal that promises those results from adding FAQ markup. Useful answers on your pages still deserve editorial attention; the agency needs to describe that work without promising a retired search feature.

How do you score three shortlisted agencies?

Use the same scorecard after each call. Score each row from 0 to 2: 0 means missing, 1 means a claim or partial evidence, and 2 means documented evidence plus a clear delivery commitment.

CriterionAgency AAgency BAgency C
Relevant buyer prompts and named competitors___ / 2___ / 2___ / 2
Named engines, baseline and measurement schedule___ / 2___ / 2___ / 2
Repeated checks and saved answer evidence___ / 2___ / 2___ / 2
On-site delivery and third-party source work___ / 2___ / 2___ / 2
Fact corrections and reporting access___ / 2___ / 2___ / 2
Itemized scope, named team and handover terms___ / 2___ / 2___ / 2
Total___ / 12___ / 12___ / 12

This is a hiring aid, not an industry benchmark. Agree any extra weighting before the calls. If your developers cannot implement changes, give implementation capacity more weight.

Do not let a high total cancel a missing measurement method. Resolve any zero for baseline or answer evidence before signing. Ask the finalist for a reference whose scope resembles yours, then confirm who handled the work and whether the reports supported decisions.

Agree the baseline before you sign

Send the finalist your approved prompts, competitor list and reporting checklist. Ask it to attach the measurement method, first delivery dates and ownership terms to the proposal. Review the baseline before approving the first round of changes.

You can also run your own baseline on Citlyze to compare with an agency's numbers. Keep the prompts, engines and dates aligned, and investigate differences before treating either figure as definitive. Start with a 14-day free trial, no card, then take the saved answers into the next agency meeting.