Start with the client decision
Small-agency fit is a hypothesis to test, not a promised outcome. Name one decision the client could make from the evidence: correct a misleading product description, improve a comparison page, or investigate a source that repeatedly appears in relevant answers.
Ask the client which buyers, market, language, and alternatives matter. Use questions from permitted sales notes, site searches, or support themes; remove personal and confidential details. A plausible prompt is not proof that buyers ask it or that AI answers influence their purchases.
If no decision would change after seeing the answers, a recurring measurement plan may be premature.
Choose buyer questions you can explain
Illustrative prompts only: these are invented examples, not customer research or measured AeoWatch results. Replace the bracketed terms using the client's approved context.
| Topic | Illustrative prompt | What to inspect |
|---|---|---|
| Discovery | Which [service category] options suit [buyer need] in [market]? | Does the client appear without being named in the question? Are the suggested alternatives relevant? |
| Comparison | How does [client] compare with [alternative] for [specific job]? | Which tradeoffs are described, and do the cited sources support them? |
| Constraints | Which [category] options meet [budget or integration constraint]? | Are the answers accurate about suitability, limitations, and availability? |
| Branded accuracy | Does [client] support [required capability]? | Is the description correct? A mention in a branded question is not evidence of unaided discovery. |
Run a small manual check first
Step 1
Fix the question set and context
Save the exact questions, client and competitor names, market, language, date, engine, model when visible, and whether web search is on. Choose a scope you can inspect within the available time and credits.
Step 2
Record each answer separately
Use a fresh session and record relevant settings. Keep the answer, mention presence, cited sources, and any refusal or error. Personalization and interface differences can still affect the result; a manual ChatGPT answer is not necessarily comparable with an API-based measurement.
Step 3
Repeat and disclose the sample basis
Ask the same questions more than once. Record usable answers, cached answers if known, and failures separately. More samples can help describe variation, but no fixed small sample count guarantees a reliable conclusion.
Step 4
Check an answer against its sources
Open the cited pages, check the claim in context, and record what the evidence can and cannot support. A citation does not establish that a source is correct or that a buyer visited it.
Step 5
Write a client decision note
Record the question, observed finding, limitation, proposed action, and when another check would be useful. Keep source material in the client's approved workspace; do not publish private reports.
Read sampled evidence without overclaiming
Illustrative arithmetic, not a customer result: if a brand appears in 3 of 5 usable answers to the same question, its observed mention rate in that sample is 60%. This does not mean 60% of buyers see it, nor does it establish a stable population rate. Five answers leave substantial uncertainty.
AeoWatch's measurement model separates mention rate from comparative share of voice and supports sample-level evidence and mention-rate confidence ranges. Check the selected prompts, engine, dates, sample size, and failures before presenting a number. A confidence range describes sampling uncertainty; it does not correct a biased question set.
Keep cached answers separate from newly collected answers. Keep partial runs and missing engines visible. Do not present an empty result or an error as zero visibility.
Compare like with like. If prompts, competitors, language, market, engine, or model change, label the new basis. A before-and-after movement alone does not prove your work caused an improvement.
Manual checking or recurring measurement?
| Topic | When it fits | What it cannot establish |
|---|---|---|
| Manual checking | You are learning the client's questions, checking one disputed answer, or deciding whether further measurement would change a decision. | A few answers do not establish a trend or the experience of all buyers. Record settings and repeated samples. |
| Recurring measurement | The client has a stable question set and a named owner who will review repeated evidence to make recurring content or positioning decisions. | Repeated runs alone do not prove usefulness, traffic, sales, or causal improvement. Budget for credits and review time. |
| Pause | The questions are speculative, the required engine is unavailable, the evidence is not inspectable, or nobody can name an action it would inform. | Buying more measurement will not resolve an undefined client decision. |
Check engine coverage before committing
Choose from the engines currently offered in your AeoWatch workspace and check the credit estimate before starting a run. Available choices depend on configuration and your workflow; an engine name in a comparison guide is not a promise that it is available in your account.
Do not promise a client Perplexity or Google AI Overviews coverage without confirming the relevant option in the workspace and inspecting a usable result from that engine. Treat a website preview and a signed-in measurement as different workflows.
This guide does not verify every engine for every account or promise a completion time. If the surface the client needs is unavailable, manual checks may still answer a limited question.
Decide whether the first result is useful
Evaluate one client job before committing to ongoing reporting.
- Can you open at least one usable answer and its measurement context rather than only a score or loading state?
- Does the question set represent the client's actual evaluation needs? Keep branded and unbranded questions separate.
- Can you explain the sample size, missing coverage, cached answers, and uncertainty in plain language?
- Can the client name a decision or next action the result supports? A viewed report is not a confirmed useful report.
- Can you deliver the review within the agreed time and credit budget without promising ranking or revenue gains?
- Is the evidence safe to share with this client? Use only authorized access and exports; never reuse another client's report as a public example.
Try the existing signup journey at current terms
Start free through the existing AeoWatch signup page. The account grant is 250 credits once per account, with unused credits expiring seven days after you first open your workspace. No payment method is needed for the grant; it does not refill monthly or renew automatically.
Signup also offers an optional seven-day Starter subscription trial. Its checkout requires a payment method, uses the remaining grant, and sets the grant expiry to the subscription trial end. It starts paid monthly billing at the checkout price unless cancelled before the trial ends. Choose Go to dashboard to skip checkout.
Published US prices checked on 8 October 2026 are Starter $29/month for 700 credits, Pro $99/month for 2,400, and Agency $249/month for 7,000. Top-up packs start at $15. Check the current price and terms shown for your market before buying; this guide changes no offer or entitlement.
Use the related resources below for measurement definitions and the homepage Pricing section for current offers. Agency fit remains unproven until the evidence helps a real client and the work is economical to deliver.