Nine red flags, in a category with no track records.

Every agency selling answer engine optimization has been doing it for roughly the same short time, which removes the usual way of judging competence. What remains is what they claim and how they talk about uncertainty.

The nine

• Guaranteed citations or placements. Nobody controls what a model says. A guarantee is either meaningless or describes manipulation.
• Confident explanations of the algorithm. Source selection is not documented. Certainty about internals indicates confidence, not knowledge.
• Reporting organic traffic as the AEO metric. That is SEO reporting with a new label on the deck.
• A default prompt set. If they have not built prompts around your buyers, they are measuring a generic market.
• Content volume as the primary deliverable. The usual binding constraints are identity and corroboration, neither of which is solved by publishing.
• No mention of entity consistency. The cheapest high-return work in the discipline. Omitting it suggests they have not done this before.
• Unwillingness to say what will not work. Every category has channels that do not pay back.
• Long lock-ins in a fast-moving field. Twelve months mostly protects them.
• Data you cannot export. If your baseline lives only in their dashboard, leaving costs you your history.

Red flags in the proposal itself

Some warning signs only show up in the paperwork. Read the proposal for these before the meeting:

• Deliverables measured in volume. “Forty AI-optimized articles per month” is a content mill with a new label, and Google's spam policies explicitly cover scaled content produced for rankings rather than readers.
• “Proprietary AI ranking signals.” Nobody outside the model companies has these.
• Review or forum “seeding”. Planting reviews or posts breaches platform rules and, for reviews, the FTC's 2024 rule banning fake reviews and testimonials.
• No baseline in the first month. Without one, nothing later can be measured.
• Termination fees or ownership clauses that leave listings, profiles or data with the agency.

What good looks like instead

Specific about method, honest about uncertainty, willing to describe the unglamorous first month, and able to show an actual before-and-after in what an assistant says about a client.

The tell is how they handle a question they cannot answer. A practitioner says they do not know and explains what they would test. Someone improvising fills the gap with confidence.

What to do if you already hired one

If you recognize several of these in your current agency, do not panic and do not cancel on the spot. First, ask for the prompt set, the raw monitoring data, and a list of every account, listing and profile they have created or changed. Those are yours and you will need them regardless.

Then ask for one thing: a before-and-after on a specific assistant answer for your business, with dates. If they can produce it, the relationship may be salvageable with a tighter scope. If they cannot, you now have the data to move, and the next agency starts from a baseline instead of from zero.

Green flags worth weighting heavily

The opposites are worth naming too. An agency that asks about your intake process, wants to hear how customers describe you, suggests a smaller first engagement than you expected, or tells you a channel is not worth it for your category is showing the behavior you want for the next year.

So is one that talks about risk unprompted: what could go wrong with a change, what could be misread, what a regulator or platform might object to. In a field this new, caution is a stronger signal of experience than confidence.

Frequently asked questions

What are the warning signs of a bad AEO agency?

Guaranteed citations, confident explanations of how models select sources, organic traffic reported as the AEO metric, default prompt sets, content volume as the main deliverable, no mention of entity consistency, long lock-ins, and data you cannot export.

Can an agency guarantee AI citations?

No. Nobody controls what a model outputs, and answers vary between sessions. A guarantee is either meaningless or implies manipulation that would not survive scrutiny.

How do I judge AEO competence when nobody has a long track record?

By method and honesty rather than tenure. Ask for a before-and-after in what an assistant says, ask what will not work for you, and watch how they handle a question they cannot answer.

Is it a red flag if an agency uses AI to write content?

Not by itself. Google's guidance is that how content is produced matters less than whether it is helpful and original. The red flag is using AI to produce volume without expertise, editing or fact-checking.

Can I get my data back from an agency?

You should be able to, and the contract should say so. Ask for prompt sets, raw monitoring results, reports and a list of every account or listing created on your behalf.