CiteGround / buyer-side / vetting checklist
How to vet a done-for-you AEO partner.
Six questions to ask any vendor selling AI-answer visibility, ours included. Every one has a checkable answer, and a vendor who squirms on more than one of them is telling you something. Take this list into every sales call; it costs the good vendors nothing.
Why does picking an AEO partner need a checklist?
AI answers are probabilistic and the field has no standards body, so results claims are cheap to make and expensive to check. Six questions separate measurement from theater.
The same prompt asked twice can return different brands, which means a vendor can truthfully screenshot a good run and quietly discard four bad ones. Nothing in this market stops them: no auditor, no standard metric, no disclosure rules. The checklist below is how a buyer builds those rules into the sales call. None of the questions require technical knowledge; all of them require the vendor to show work instead of describing it.
What baseline sits behind that percentage claim?
Ask for the absolute numbers and the denominator. A '+61% visibility' claim means nothing without the from-what, over-how-many-prompts, measured-how. No baseline, no claim.
Percentage lifts are the industry's favorite unit because they hide the base. Referrals up 29% in a month: from 7 to 9? Visibility up 61%: on which prompt set, run how many times, on which engines, measured by whose tool? A serious vendor answers with the absolute before-and-after, the prompt count, the run count, and the measurement method, and does not mind being asked. A vendor who answers with another percentage has answered.
Are the credit terms on a decision clock?
Some vendors credit the diagnostic fee only if you sign within days. That converts thinking time into sunk cost. Get credit terms in writing with no deadline attached.
A common structure: pay a few thousand for a diagnostic, and the fee credits against the retainer, but only if you sign inside a two-week window. Read that back slowly: deliberating past the deadline costs you the full diagnostic fee. Whatever the intent, the mechanics are a decision clock: credits are fine, clocks on credits are pressure. Ask for the credit terms in writing and check whether a deadline appears anywhere in them.
Are community mentions disclosed or seeded?
Reddit and forums ground many AI answers, so vendors buy or manufacture mentions there. Seeded mentions carry the platform and brand risk on YOUR name. Ask how, exactly.
Community threads are real grounding sources: in our own category reading, reddit appears in the citations of half the measured prompts. That makes "we get you mentioned on Reddit" a genuinely valuable line and an ethically loaded one. Undisclosed seeded mentions violate platform rules, and when they get flagged, the account that burns is your brand's, not the agency's. The question to ask is one word deep: how? Good answers involve disclosure and posting as yourself. Vague answers involving "our network" are the thing to walk away from.
Can they show you one raw archived run?
Dashboards summarize; a measurement practice can produce the run itself: the prompt, the engine, the timestamp, the full answer, the citations. Ask them to re-run one live.
Every vendor shows charts. Very few can produce the underlying object: this prompt, this engine, this timestamp, this full answer text, these cited pages, archived at the time and never edited. If the raw run exists, asking for one costs the vendor thirty seconds. If it does not exist, the charts are summarizing something you cannot inspect. The strongest version of this test is live: ask them to re-run any prompt from their own report on the call and watch whether the number survives.
What happens contractually when nothing moves?
Ask what the contract says at week 4 and at renewal. Good answers name a metric ladder and a consequence. Bad answers describe effort.
AI-answer work has real lag, so no honest vendor promises a share-of-voice number by a date. The dishonest move is using that lag to make everything unfalsifiable. The middle path is a metric ladder with contractual teeth: placements live by a named week, then presence in cited sources, then answer share, with the renewal decision tied to position on that ladder and a stated consequence when the vendor's own deliverables slip. Ask where that language lives in their contract. "We work closely with you" is not a location.
How does CiteGround answer these six?
In writing, on this site: denominators on every number, credits with no deadline, disclosed-only community work, a verbatim archived run in every sample report, week-4 contract language.
Fair is fair; here are ours. And yes, we wrote a rubric we happen to pass; the honest defense is that every answer below is checkable rather than asserted, so you never have to take the rubric's author at his word. Every number we publish states its denominator and variance (methods). The roadmap credit has no decision clock. Community answers are posted as ourselves, disclosed, or not at all. Every report reproduces one archived answer verbatim with its citations, and any prompt in it can be re-run live on a call (see the sample report). The retainer contract names the week-4 interventions-live clause, with the term extending day for day if we miss it, and renewal rides on ladder position. If a vendor you like answers all six better than that, hire them; the checklist did its job.