The claim: technology makes predictable growth easy
Some AI-assisted marketing services perform useful work. Others describe a limited activity in language that suggests a much broader business outcome. Buyers need a repeatable evaluation method that does not depend on liking a sales presentation or distrusting every new tool.
ZeroDark's approach is to reconstruct the promise as a measurable statement. Identify the buyer, result, unit, starting conditions, time window and evidence. A provider should be able to explain each without changing definitions halfway through the conversation.
Twelve questions to put in the buying brief
- What exact outcome is promised, and what event qualifies one unit?
- Which parts are contractually guaranteed, forecast or merely illustrated?
- For search, what queries, markets and result types are included?
- What evidence shows that the chosen queries represent relevant buyers?
- For AI visibility, can I inspect the full prompt set and unfavorable answers?
- For leads, are we buying contacts, replies, meetings or accepted opportunities?
- For revenue, is the number contracted, invoiced, collected or normalized recurring revenue?
- What starting assets, budget, staffing and buyer actions does the result assume?
- How were case studies selected, and what comparable-outcome evidence exists?
- What costs, exclusions, eligibility conditions and remedies apply?
- Which source records can be inspected, and how are errors challenged?
- What would count as a reason to change or stop the program?
These are decision questions, not an accusation against a particular company. General advertising guidance, search-platform documentation and AI evaluation frameworks support careful substantiation and testing. They do not allow a buyer to infer misconduct from an unanswered question alone. FTC: advertising guidance for small business.
Use an evidence register rather than a confidence score
| Register field | Example entry |
|---|---|
| Exact promise | Ten attended meetings with agreed account-fit criteria |
| Source | Proposal section, version and date |
| Measurement | Attendance record and sales acceptance log |
| Evidence status | Demonstrated, partial, unavailable or contradicted |
| Open question | Treatment of cancellations and existing accounts |
| Decision | Proceed to a bounded pilot after the definition is settled |
Avoid averaging these statuses into an impressive percentage. One unresolved condition may be commercially decisive even if many minor fields are complete. The purpose is to expose the missing evidence, not produce another proprietary-looking score that cannot be explained.
A useful pilot tests uncertainty
Choose a representative segment, preserve starting evidence and state the acceptance criteria before delivery. Inspect a sample early enough to correct definitions and collection problems. Keep activity measures separate from outcome measures and report the full eligible cohort.
For a search pilot, that may mean a fixed set of relevant non-brand queries and destination pages. For lead generation, it may mean a verified sample with a documented rejection process. For AI assistance, it may mean error rates, review time and the quality of a completed task. A pilot should reduce a specific uncertainty, not simply generate a favorable screenshot.
Turn findings into a decision
A provider might demonstrate a useful activity but lack evidence for a broad revenue forecast. Buying the activity on clear terms can still be reasonable. Another provider may offer an impressive guarantee with conditions that do not fit your market. Rejecting that fit does not require declaring the company illegitimate.
ZeroDark positions marketing intelligence between the sales promise and the business decision. The work is to uncover what is actually happening, understand its significance and execute an improvement that can be checked. The same discipline should apply to our own audit, SharkBlade platform and expert-team proposals.
Supporting documentation
FTC: advertising guidance for small businessGoogle: hiring an SEONIST: AI Risk Management Framework