Direct answer
Choose an AEO agency the same way you would hire any specialist: make them prove their process before you sign anything. A proof-first selection process means asking for real examples of work, transparent reporting you can read without a translator, and a clear explanation of what they will actually change on your site. Good agencies show their method. Weak ones sell outcomes they cannot control. Your job is to separate the two before money changes hands.
What should a proof-first selection process look like?
Start with evidence, not promises. Ask the agency to walk you through a recent project: what the site looked like before, what they changed, and how they measured the effect. You want to see the actual pages, the schema they added, the content they rewrote, and the reporting they delivered. If everything is under NDA and they cannot show a single anonymized example, that is a signal worth noticing.
Next, review how they report. Strong reporting ties activity to observable signals: which pages earned answer-box style visibility, which queries surfaced your brand in AI summaries, how organic and referral traffic moved, and how those trends line up with the work they did. Reporting should distinguish between what they influenced and what simply happened. Vague dashboards full of vanity metrics are easy to produce and hard to trust.
Then look at process. A serious agency will usually begin with an SEO audit to understand crawlability, structure, and existing content quality before proposing anything. From there, thoughtful AEO service planning should map your priority questions, the pages that answer them, and the technical and editorial work required. If an agency skips discovery and jumps to a fixed package, they are guessing.
Finally, talk to references. Ask past clients whether the reporting matched reality and whether the agency was honest when results were slow.
What can an AEO agency influence versus not control?
An agency can influence a lot. They can improve the clarity and structure of your content so machines and people can parse it. They can add and validate schema, tighten internal linking, fix crawlability issues, build focused answer content for real questions your audience asks, and strengthen author and business signals that support trust. They can also improve site speed and mobile usability, which affect how well your pages get read and rendered.
What they cannot control is how any given AI system or search engine chooses to display, cite, or summarize your content on a specific day. AI-search features change frequently, models get retrained, and answer formats shift without notice. No honest agency can promise a spot in an AI overview, a fixed ranking, or a guaranteed citation. They can raise the probability that your content is eligible and worth surfacing. They cannot force the result.
Be cautious of anyone who blurs that line. Influence over inputs is real work. Control over outputs is a claim no one can back.
What Google says about content for AI-search features
Google has been consistent: create helpful, reliable, people-first content, and make it crawlable. Its guidance for AI features points back to the same fundamentals as its core content advice. Content should be written for people first, demonstrate real experience and expertise, and answer questions clearly. Pages must be technically accessible to Googlebot so they can be crawled and indexed in the first place. Google explicitly says there is no special markup or trick that guarantees inclusion in AI experiences. The path is quality content that satisfies the person asking.
That aligns neatly with a proof-first agency. If a partner's plan is built on genuinely helpful, well-structured, crawlable content and honest measurement, they are working with how these systems are designed to behave rather than against them.
Red flags to watch: guaranteed rankings or placements, refusal to show methodology, reporting you cannot verify, promises of instant results, and any claim of a secret formula. Trustworthy agencies sell a process and prove it. Everyone else is selling hope.
How Do You Build a Vendor Scorecard That Actually Predicts Fit?
Before you sign anything, turn your gut feeling into a scored sheet. Rate each agency from 1 to 5 across seven weighted areas, then compare finalists side by side instead of relying on the pitch that felt best.
| Category | Weight | What a 5 looks like |
|---|---|---|
| Proof of past work | 25% | Named clients, before/after snapshots, retained access to reporting |
| Reporting transparency | 20% | Raw data access, defined metrics, no vanity numbers |
| Technical competence | 20% | Talks in schema, crawlability, and structured data, not buzzwords |
| Content quality | 15% | Writers with subject depth, real editing, human review |
| Measurement honesty | 10% | Explains what they can and cannot influence |
| Communication cadence | 5% | Named contact, agreed check-in rhythm |
| Contract flexibility | 5% | Reasonable exit terms, you keep the assets |
Anything scoring under 3 in proof, reporting, or measurement honesty should give you pause no matter how strong the rest looks.
What Evidence Should You Demand Before Signing?
Ask every finalist to prove their claims with artifacts, not adjectives. Run these evidence tests:
- The named-client test. Request two or three clients you can actually look up. Vague references to "a national brand" that cannot be verified are worth nothing.
- The raw-data test. Ask to see a real reporting dashboard or export, with the logos redacted if needed. You want to see how they track visibility in AI answers, branded query trends, citation mentions, and referral traffic, not a slide of hand-picked wins.
- The methodology test. Have them walk you through how they measure change over time. If they cannot describe their tracking method in plain language, they probably do not have one.
- The sample-work test. Read a piece of content they produced. Check whether it reads like a knowledgeable human wrote it or like generic filler.
- The attribution test. Ask how they separate their impact from seasonality, brand growth, and other channels. Honest answers acknowledge overlap.
Be direct about a hard truth: no one controls whether an AI assistant cites you. A credible agency improves the structured, authoritative, well-organized signals that make your content easier to surface, and they say so plainly.
Green flags worth rewarding on your scorecard:
- They tell you what they cannot promise before you ask.
- They give you access to and ownership of your own data and content.
- They tie work to observable signals like structured data coverage, entity clarity, and citation trends.
- They set baselines first and report against them.
- They show comfort with your CMS, schema, and analytics stack.
Red flags that should drop a score fast:
- Guarantees of a top spot in AI answers or a fixed number of citations.
- Refusal to name any verifiable client or show any real reporting.
- Reporting built only on traffic screenshots with no method behind them.
- Content produced with no human editing or subject expertise.
- Long lock-in contracts where you lose access to everything if you leave.
- Talk of "secret algorithms" or proprietary tricks they will not explain.
Which Questions Expose Weak Technical and Content Plans?
When an agency proposes specific work, press on the details. For technical work, ask:
- Which structured data types will you add or fix, and how will you validate them?
- How will you audit crawlability and rendering for AI systems and search engines?
- How do you handle entity consistency across our site, profiles, and citations?
- What will you measure before and after each technical change?
For content work, ask:
- Who writes and edits, and what subject expertise do they bring?
- How do you research questions our audience actually asks?
- How do you keep content accurate, current, and free of unsupported claims?
- How will we know a piece performed, and over what timeframe?
The strongest agencies answer these without flinching, admit uncertainty where it exists, and hand you the evidence to check their work yourself. Score accordingly, and the right partner tends to separate from the rest.
What Should You Expect After You Sign?
The agency selection process does not end at the proposal. What happens in the first month tells you more than any pitch deck. Set clear expectations for reporting before you commit, and put the ownership terms in writing.
Reporting that survives scrutiny
Ask for a sample report before you sign, not after. A useful AEO report shows the questions your buyers actually ask, which answer engines surface your brand, and where your content appears in cited sources. Screenshots of a single lucky citation are not reporting. Look for a repeatable method: the same prompts run on a schedule, tracked over weeks, with the raw data attached so you can check the work yourself.
Reporting should separate what the agency controls from what it does not. Content published, schema deployed, and pages indexed are inputs the agency owns. Citation frequency and answer-engine visibility are outcomes influenced by many factors. An honest report labels both and never blends them into one vanity number.
Contract and ownership questions
Read the contract for who owns the work. You should own the content, the schema markup, the structured data, and any accounts created on your behalf. If the agency retains the content or hosts it on a domain they control, you lose everything when the relationship ends. Confirm this in the agreement, not in a sales call.
Ask these before signing:
- Do I own all published content, code, and structured data outright?
- What is the notice period, and what happens to in-progress work if I leave?
- Are the analytics, Search Console, and CMS accounts registered under my email?
- Which deliverables are guaranteed as work product, and which are described as attempts?
If an answer is vague, get it in writing or treat it as a red flag.
A 30-day evaluation
Structure the first month as a trial you can measure. Agree on a short list of deliverables with dates: a content audit, a set of answer-ready pages, schema validation, and a baseline report of current answer-engine visibility. At day 30, you are not judging outcomes yet. You are judging execution. Did the work ship on time? Is it published on your domain under your control? Does the reporting match what was promised?
Use the the research hub to compare their method against documented practice, and review a a case-study route to see how a full engagement is structured. A capable partner welcomes this kind of scrutiny and gives you the raw data to verify their claims.
Five questions every AEO agency should answer
Use the same questions with every finalist so the comparison is fair and the decision is not driven by the best sales presentation.
- What specific technical, content, and measurement work will you deliver in the first 30 days?
- Which inputs can you influence, and which search or AI outcomes can you not promise?
- How will you establish a baseline, preserve raw data, and make each report reproducible?
- Who owns the published content, structured data, analytics access, and accounts if the relationship ends?
- Which assumptions about our audience, site, or market still need to be tested before work begins?
How SCALZ.AI reviews an AEO engagement
At SCALZ.AI, we start by documenting the facts a client can verify: the site’s current technical condition, the pages that already serve a real audience, the questions worth answering, and the conversion data available for measurement. We then separate proposed work from outcomes no agency can control. A plan should name the page or system being improved, the evidence used to prioritize it, the owner, and the way the change will be checked after release.
That approach is deliberately less dramatic than a promise of instant AI visibility. It gives the client a usable record of what was changed and why, while leaving room to revise the hypothesis when the evidence says to. The goal is an accountable engagement, not a claim that an answer engine will behave on command.
We keep the decision trail usable for the people who inherit the work: the stated baseline, the approved changes, the checks run after release, and the questions that remain unanswered. That record makes a client less dependent on a vendor’s interpretation of a dashboard.
