Our position: a client report fails when it is technically impressive but leaves the room unsure what to approve.
What you should leave with
- Lead with buyer decisions, not model names.
- Show excerpts beside classifications.
- Label observation and inference differently.
- End every finding with an owner and retest.
What should the executive summary say first?
State the material gap in one sentence: how often the client reached the shortlist on important questions, who displaced them, and which repeated evidence problem deserves action. Add scope and confidence immediately below it.
Executives need the consequence before the mechanics. ‘The brand was absent from seven of ten implementation-focused questions while two competitors were repeatedly justified by integration proof’ is more useful than ‘Gemini visibility fell eight points.’ It connects the result to a buyer decision.
Use three findings at most on the opening page. Rank them by commercial value, frequency, and fixability. A long list treats a one-off answer error as equal to a repeated omission across the client's core category.
- Scope and confidence
- Largest valuable recommendation gap
- Repeated competitor or source pattern
- One decision requested from the client
Evidence used in this section
How much prompt evidence should clients see?
Show enough evidence to verify every material claim: the exact prompt, answer excerpt, platform, date, recommended brands, source links, and reviewer note. Keep full answers in an appendix or drill-down view.
Do not ask clients to trust a colored dot. Put the text that triggered the classification beside the label and highlight the recommendation reason. When an entity match is ambiguous, show why the reviewer accepted or rejected it.
Source visibility needs careful wording. A linked page is observed evidence; an uncited explanation is not proof of model influence. Clients can understand this distinction when it is stated plainly and consistently.
- Prompt wording and buyer context
- Verbatim evidence excerpt
- Citation ownership and source type
- Reviewer confidence and notes
Evidence used in this section
How do you structure the client meeting?
Use a 30-minute sequence: align on scope, explain the top three lost decisions, inspect one complete example, agree on the cause hierarchy, and approve the first fix sprint. Send the evidence appendix after the decision is clear.
Choose one example that represents the pattern, not the most dramatic answer. Walk from prompt to recommendation, source environment, evidence difference, proposed action, and retest. That teaches the method while keeping the discussion grounded in the client's market.
Invite disagreement at the classification and business-value layers. The client may know that a named competitor is irrelevant or a prompt reflects a low-value segment. Correct the model of the market, preserve the audit history, and document the scope change.
- STEP 1
Orient
Confirm market, platforms, prompt coverage, and confidence before showing scores.
- STEP 2
Prioritize
Explain the three highest-value patterns and why they outrank the rest.
- STEP 3
Inspect
Open one complete prompt, answer, source set, diagnosis, and proposed fix.
- STEP 4
Decide
Approve owners, deliverables, acceptance criteria, and the next retest date.
Evidence used in this section
Which numbers belong in a recurring client report?
Report stable prompt coverage, valuable recommendation rate, competitor share, cited-domain changes, factual error status, and task impact. Pair every trend with the prompts that changed and an annotation for method changes.
A rising average can hide a loss on the most valuable prompt family. Segment by decision type, buyer, or region and keep the stable core separate from exploratory prompts. The client should see whether broad progress occurred or one volatile answer moved the chart.
Use status language for implementation: published, crawled, third-party corrected, observed in answer, and persistent on repeat. This prevents completed production work from being mistaken for a guaranteed platform outcome.
| View | Question answered | Required context |
|---|---|---|
| Executive trend | Are valuable outcomes improving? | Stable scope and annotations |
| Prompt changes | Exactly what was won or lost? | Answer excerpt and repeat count |
| Task impact | Did the intended signal appear? | Ship, crawl, and retest dates |
Evidence used in this section
What makes a client report misleading?
A report becomes misleading when it hides scope changes, treats citations as recommendations, reports unstable single runs as trends, or implies that completed agency work guarantees model behavior. Show the boundary beside the result, not in tiny footer text.
Avoid vanity density. Twenty charts can create the impression of certainty while concealing a small prompt set. Choose the few visualizations that support a decision and provide the raw evidence for stakeholders who want to audit the conclusion.
Do not bury bad news. Lost mentions, newly cited misinformation, and failed tests are part of the monitoring product. Reporting them promptly builds more trust than protecting a smooth score line.
- No silent prompt replacement
- No unqualified causal language
- No score without numerator and denominator
- No hiding regressions behind net averages
Method boundary: If methodology changes materially, mark a new baseline. Connecting incompatible periods with one continuous line creates a trend that did not actually exist.
Evidence used in this section
Questions that change the decision
Frequently asked questions
Should the report show all AI platforms together?
Use a combined executive view only when the weighting is disclosed, then provide platform-level results. Different source systems and answer behavior can make the aggregate hide important differences.
How do we explain a low-confidence finding?
State why confidence is low, such as one run, ambiguous entity matching, or missing source exposure, and define the repeat needed before action.
Should raw AI answers go in the main deck?
Use short excerpts in the main story and keep complete answers in a linked evidence appendix. Material claims should always be traceable to the full response.
What should the final slide contain?
List the first fix sprint with finding IDs, owners, deliverables, expected signals, dependencies, and the scheduled retest.
Primary sources and research
Platform documentation supports factual statements. Where we describe an audit method or prioritization rule, that is AnswerMentions' operating judgment and is labeled as such.
- [1]NIST: AI Risk Management FrameworkNIST frames AI risk work around governance, mapping, measurement, and management, a useful model for separating observations from decisions.
- [2]FTC: advertising and marketing basicsThe FTC states that advertising claims must be truthful, non-deceptive, and supported by evidence when appropriate.
- [3]OpenAI Help: accuracy and citationsOpenAI warns that ChatGPT can produce incorrect facts and fabricated references, so consequential claims should be checked against reliable sources.
- [4]Perplexity Help Center: how sources workPerplexity explains that it searches the web, identifies sources, and synthesizes an answer with citations, making source inspection central to evaluation.
- [5]Aggarwal et al.: Generative Engine OptimizationThe KDD 2024 paper evaluates generative-engine visibility in a controlled benchmark; it is evidence that visibility can be studied, not a universal ranking recipe.
- [6]Google Search Console: performance report documentationSearch Console documents query, page, country, and device dimensions, which are useful supporting signals but do not identify every AI recommendation exposure.