To evaluate an AI marketing service, ask 12 questions in four areas: accountability (who owns the result), humans (who reviews the AI's output), transparency and data (what you can see, own and keep private), and economics (what the price buys and how it grows). Good answers contain names, numbers and processes. Red flags are adjectives.
Every marketing service in 2026 claims AI, so the label no longer tells you anything. This guide gives all 12 questions in one table, then explains each one: what a good answer sounds like, and the red flag that should end the meeting. The checklist works on any provider, whether or not you ever talk to us.
Why does evaluating an AI marketing service matter now?
Because adoption is everywhere and results are not. 58% of US small businesses say they use generative AI, up from 40% in 2024 and 23% in 2023, according to the U.S. Chamber of Commerce's Empowering Small Business report (2025). 91% of SMBs with AI say it boosts their revenue, according to Salesforce's SMB Trends Report (2024; self-reported by 3,350 SMB leaders). Yet only 14% of small business owners say AI is fully embedded in their core operations, according to Goldman Sachs' 10,000 Small Businesses Voices survey (2026; a survey of program alumni, not a national sample).
Two services can quote the same price, claim the same AI, and produce very different results, because the difference lives in structure: who owns outcomes, who reviews the work, what you can see, and how the incentives point. None of that appears in a sales deck, which is why you have to ask.
The 12 questions at a glance
| # | Question | Good answer | Red flag |
|---|---|---|---|
| 1 | Who owns the outcome? | A business metric and a named person | Outputs: posts, impressions, launches |
| 2 | What happens when a metric stalls? | A documented diagnose-and-reallocate process | "We'd discuss it on the monthly call" |
| 3 | What are the cancellation terms? | Month-to-month, written notice period, clean handover | 6–12 month lock-in, auto-renewal, exit fees |
| 4 | Who reviews AI output before it ships? | Names, roles, and where they sit in the workflow | "Our team reviews everything" |
| 5 | How much senior time is in my tier? | Hours or cadence per tier, in writing | Same "full oversight" at every price |
| 6 | What does the AI do, and what do humans decide? | A specific line between the two | "Proprietary AI does it all" |
| 7 | Do I see work before it ships? | Approval flow with defined turnaround | No approval path, or you approve everything |
| 8 | Live metrics or monthly slides? | A dashboard you can open any day | A monthly PDF as the only window |
| 9 | Do I own the accounts, assets and data? | Yes, all of it, from day one | Accounts under the provider's umbrella |
| 10 | Is my company data used to train AI models? | A written no, a vendor list, deletion on exit | "We use the leading models" |
| 11 | What does the price include, and what costs extra? | A published scope per tier; ad spend paid by you | Scope by request; percentage-of-spend fees |
| 12 | How does pricing scale as I grow? | Published tiers and thresholds | "We'll figure it out as you grow" |
Accountability: who answers when the number stalls?
These three questions establish whether anyone is on the hook for results. They come first because every other answer is decoration if nobody owns the number.
1. Who owns the outcome? AI can produce infinite activity, which makes it easier than ever for a provider to look busy while nothing moves. A good answer names a metric and a person: leads, qualified pipeline, CAC, and who is on the hook for it. Red flag: the outcome named is an output, such as posts published, impressions served, or campaigns launched.
2. What happens when a metric stalls? Every growth program hits a plateau; what distinguishes providers is the documented response. A good answer describes a process: diagnose within a defined window, present the analysis, reallocate effort. Red flag: "we would discuss it on the next monthly call." A stall discovered in week one and addressed in week five costs you a month of spend.
3. What are the cancellation terms? Commitment terms reveal how a provider expects to retain you. A good answer is month-to-month by default, with the notice period, the refund rule for any prepaid term, and the handover of work in progress all written down before you sign. Red flag: a 6–12 month minimum justified by "results take time," auto-renewal buried in the contract, or an exit fee. Results do take time, but a provider confident in month three should not need you locked in through month nine to prove it.
Humans: who is actually behind the AI?
AI is now table stakes; the human layer is where services differ. These three questions find out whether the humans are real, senior, and structurally involved, or a photo on the website.
4. Who reviews AI output before it ships, by name and role? Unreviewed AI output is where generic content, factual errors, and off-brand claims come from, and the quality risk is not only reputational. Google's guidance on generative AI content says that "using generative AI tools or other similar tools to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse," and tells site owners to "focus on accuracy, quality, and relevance" (Google Search Central, updated December 2025). A good answer gives you actual names, roles, and where they sit in the workflow, the way we list our team on our about page. Red flag: "our team reviews everything" with no names, or review that turns out to be a spelling pass, not judgment about whether the work is right.
5. How much senior time is in my tier? Cheaper tiers run thinner human review; the honest question is how much thinner. A good answer quantifies the difference between tiers in strategist attention and review depth, in writing. Red flag: every tier claims the same "full senior oversight" at wildly different prices. One of those tiers is mispriced, and it is not the expensive one.
6. What does the AI do, and what do humans decide? A provider who cannot draw this line crisply either does not know or does not want you to. A good answer is specific: AI drafts, analyzes, and produces variants; humans set strategy, approve what ships, and decide what changes. Red flag: vague gestures at proprietary AI doing everything, or, equally, AI as garnish on a traditional agency's manual work sold at AI-era prices. Where that line sits today is the subject of can AI replace a marketing agency?
Transparency and data: what do you see, own and keep private?
These four questions test whether you can verify what the provider claims without asking permission, and whether your assets and data stay yours. The pattern to look for is visibility and ownership by default, not by request.
7. Do I see work before it ships? You are accountable for what goes out under your brand, whoever produced it. A good answer describes an approval flow with defined turnaround, plus a sensible default for routine items. Red flag: publishing without any approval path, or its mirror image, where everything requires your sign-off and you quietly become the bottleneck they will later blame.
8. Live metrics or monthly slides? Reporting cadence is the difference between steering and archaeology. A good answer is a dashboard you can open any day, showing work shipped and results attributed, the standard we hold ourselves to with the Scale AI-hub. Red flag: a polished monthly PDF as the only window, which gives a provider four weeks of cover per underperforming decision.
9. Do I own the accounts, the assets and the data? Ad accounts, analytics, domains, content, audience lists, and the performance history inside them should all live under your ownership, with the provider as an invited user. A good answer is an unhesitating yes, all of it, from day one, plus an export of anything held in the provider's own systems when you leave. Red flag: accounts run under an agency umbrella you would lose by leaving. That converts switching costs into a retention strategy, and prices your exit into every month you stay.
10. Is my company data used to train AI models? An AI service feeds your customer lists, pricing and strategy into software, so ask where it goes. A good answer is a written statement naming the AI vendors that process your data, confirming that it is not used to train models (theirs or a third party's), offering a data processing agreement, and describing deletion when you leave. Red flag: "we use the leading models" with no vendor list, or staff pasting your customer data into consumer chatbots on personal accounts.
Economics: what does the price actually buy?
The final two questions price the relationship over its whole life, not just the first invoice. Cheap entry pricing with opaque scaling is the oldest trick in services.
11. What does the price include, what costs extra, and is ad spend mine? AI-era services vary wildly in what a monthly fee covers. A good answer is a published list per tier (channels, volume, strategy cadence) like the one on our pricing page, with extras listed in writing before you sign and ad budget flowing from your card to your accounts. Red flag: scope discovered by request, where every second ask turns out to be an add-on, or ad spend routed through the provider with percentage-of-spend pricing, which rewards them for spending more of your money, not better. What subscription pricing in this category usually covers is broken down in Growth-as-a-Service pricing.
12. How does pricing scale as I grow? You are not just buying this month; you are buying the price curve. A good answer ties price to defined tiers with published thresholds, so you know today what doubling volume costs. Red flag: "we will figure it out as you grow," which reliably means renegotiation from the weak side of the table, after your data and workflows already live inside their system.
How do you run the evaluation?
Send the 12 questions in writing before any demo, and hold answers to a simple bar: names, numbers, and processes are answers; adjectives are not. On the call, drill where the written answer was thinnest. Ask for evidence over claims: real client examples with attributable results, and references you can call. A provider too new to have them should say so plainly. We are one: Scalehackerlab was founded in May 2025, and what we publish under growth playbooks are illustrative 90-day plans, labeled as such, not client results. Weigh how the provider handles the questions themselves; one that welcomes scrutiny will behave the same way when a campaign stalls.
A simple scoring method keeps the comparison honest: two points for a specific written answer, one for a partial one, zero for adjectives. Score all four groups, and do not let averages hide a gap: a high total with a zeroed group is a structural no, because the groups do not compensate for each other. A cheap service that fails accountability is not cheap. How the categories trade off against agencies and platforms is a decision we mapped in agency vs. platform vs. GaaS.
Where to start
Start with whoever you already pay: your current agency, tool stack, or freelancer bench, scored against all 12 before you evaluate anyone new. That baseline turns vendor selection from a sales conversation into a comparison. Expect the exercise to disqualify most of a shortlist. That is the point: the cost of a wrong choice is not the fee, it is quarters of stalled pipeline with a switching cost at the end.
For our part, Scalehackerlab publishes its own answers (team, tiers, ownership, and what is included) on /about, /faq and /pricing, and the model behind them is explained in what is Growth-as-a-Service. If you want a second opinion on where your current setup stands, the free growth assessment returns a strategy document in 48 hours, no credit card required.