Our position: a monthly report earns its fee by explaining what changed and what to do, not by refreshing the same score.
What you should leave with
- Keep a stable core and separate exploration.
- Repeat material gains and losses.
- Annotate method and business changes.
- Carry unfinished findings into owned tasks.
Which sections belong in every monthly report?
Include an executive change summary, stable-core scorecard, prompt gains and losses, competitor movement, source changes, factual error status, completed work, and next-month priorities. Every trend should link to its underlying answer evidence.
Lead with material changes, not all changes. A new recommendation on a high-value evaluation prompt matters more than five casual mentions on educational questions. Explain the buyer decision, competitor displacement, source change, and whether the result repeated.
Include a no-change statement when appropriate. If the team shipped work but no persistent answer outcome has appeared, say so and report intermediate signals such as publication, crawl, directory approval, or new third-party coverage.
- What changed and why it matters
- Which evidence supports the change
- What work shipped and its current signal
- What the team will do next
Evidence used in this section
How do you preserve a comparable baseline?
Rerun a versioned core with the same wording, platform scope, market, classification rules, and weights. Put new questions in an exploratory set until they are intentionally promoted into a new baseline version.
Business reality still changes. Products launch, competitors enter, locations close, and platforms alter search behavior. Annotate these events and create a baseline break when they materially change the test; do not force incomparable periods into one smooth line.
Keep raw runs even when an answer appears anomalous. Repeat the result and label its stability. Deleting inconvenient outcomes makes the series cleaner at the cost of making it untrustworthy.
- Versioned prompt inventory
- Fixed classification policy
- Documented platform and market context
- Visible baseline breaks and anomalies
Evidence used in this section
How should a monthly review investigate change?
Start with prompts that crossed a meaningful state, inspect complete answers and source sets, compare competitor evidence, check recent client work, and retest before assigning cause. Correlation is a lead, not a victory claim.
When a client becomes recommended, ask what reason the answer gives and whether new sources support it. A published page, directory correction, review change, or unrelated platform update may coincide. Use precise language: observed after, consistent with, or not yet attributable.
For regressions, look for newly cited competitor pages, stale client facts, source disappearance, access issues, and changed recommendation framing. A lost mention can reveal a source-maintenance task as often as a need for new content.
- STEP 1
Detect
Flag valuable state changes, source shifts, errors, and competitor movement.
- STEP 2
Inspect
Compare full answers, reasons, citations, context, and recent implementation events.
- STEP 3
Repeat
Rerun material changes and classify their stability before reporting a trend.
- STEP 4
Act
Close, revise, or create tasks with evidence IDs and next retest criteria.
Evidence used in this section
Which monthly metrics are decision-grade?
Use valuable recommendation coverage, share of voice on the fixed set, persistent gains and losses, source-domain movement, verified error status, and task-to-signal progression. Always show denominators and scope.
A net score can hide churn. Report gross wins and gross losses, then explain whether they cluster by buyer, platform, product, region, or evidence type. One stable gain in a core segment may outweigh noisy movement elsewhere.
Compare implementation progress with observed outcomes without collapsing them. ‘Published’ and ‘recommended’ are different states; the report should make the distance visible so teams do not over-credit work or abandon it prematurely.
| Metric | Decision it supports | Stability rule |
|---|---|---|
| Core recommendation rate | Overall shortlist movement | Same denominator and weights |
| Persistent changes | Where to investigate or defend | Repeated before escalation |
| Task-to-signal status | What implementation unlocked | Separate ship, crawl, source, and answer |
Evidence used in this section
What creates false movement in monthly reporting?
Silent prompt edits, changing competitor sets, one-run observations, altered location context, classifier changes, and platform-access differences can all create artificial trends. Record these as method changes before interpreting performance.
Do not smooth volatility by quietly swapping difficult prompts. Keep the stable set and explain why an exploratory question may replace it at the next formal version. A transparent break is more useful than a continuous but synthetic history.
Monthly does not mean every task needs a new article. Maintain accurate profiles, update evidence, repair technical access, and improve existing decision pages when those actions address the finding more directly.
- No silent method changes
- No single-run victory laps
- No net score without gross movement
- No automatic content quota
Method boundary: Answer engines are not deterministic rank trackers. Monthly monitoring reduces uncertainty through consistent observation; it does not remove variability.
Evidence used in this section
Questions that change the decision
Frequently asked questions
Is monthly reporting frequent enough?
Monthly suits most implementation programs. Weekly sampling can help volatile or high-value categories, while quarterly may be enough for slow markets with little active work.
Should every prompt run more than once?
Repeat high-value and materially changed prompts first. Running every low-value prompt many times may add cost without changing the decision.
How do we report a platform redesign or model change?
Annotate the date and context, test for a structural break, and establish a new baseline when results are no longer comparable under the old method.
What if nothing improved this month?
Report the truth, show intermediate implementation signals, explain blockers, and change the fix hypothesis only when evidence supports it.
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]OpenAI Help: accuracy and citationsOpenAI warns that ChatGPT can produce incorrect facts and fabricated references, so consequential claims should be checked against reliable sources.
- [3]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.
- [4]OpenAI: ChatGPT searchOpenAI describes ChatGPT search as providing timely web answers with links to relevant sources and publisher content.
- [5]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.
- [6]Google Search Central: spam policiesGoogle treats scaled pages made primarily to manipulate rankings as abuse, regardless of whether automation, people, or both produced them.