Investigate suspicious engagement by recording what you can observe, testing alternative explanations, and asking the creator for specific context. Treat repeated comments, unusual ratios, and sudden changes as reasons to review. None identifies who arranged the activity or proves that the creator bought it. A fair fake influencer engagement audit ends with a hiring decision and its evidence limits, rather than an unsupported accusation.
The worksheet below separates an observation from an inference. Its escalation rules are editorial recommendations for campaign screening. YouTube examples explain platform mechanics; they should not be treated as rules for every social network.
Write the observation before naming the cause
Start with a sentence another reviewer could verify from the same material:
In this hypothetical sample, 14 of 40 visible comments repeat the same two-word phrase across three videos.
That describes the sample. Calling those comments purchased adds a claim about their origin that the observation does not establish.
Modash's engagement-rate guide recommends inspecting comments across posts rather than relying on a rate alone. That is a useful review step. Repeated commenters or emoji replies still require context before you assign a cause.
Possible explanations to test include a creator asking viewers to repeat a phrase, a recurring community joke, or unsolicited spam. These are hypotheses, not automatic excuses. Look for the prompt in the content, dates that match the explanation, and whether the same pattern appears elsewhere.
Platform rules concern conduct. YouTube prohibits artificial metric inflation, while allowing creators to ask viewers to like, share, subscribe, or comment. The existence of a comment prompt therefore does not itself establish a violation.
Keep a sample record that exposes its limits
Before opening individual threads, set the period, content format, post-selection rule, and comment-selection rule. Use the same rule for every candidate. For a broader content review, choose a fixed recent-post sample before investigating exceptions.
A practical starting rule might be six recent videos in the format you would commission, with up to 20 visible top-level comments per video. This is a suggested workload limit, not a statistically validated fraud test. Record fewer comments when fewer are available. Keep creator replies separate so a conversation does not inflate your count of audience responses.
For each post, save:
- Its URL, publication date, format, and your observation time.
- The visible count and number of comments you inspected.
- The sort order, exclusions, and any loading limit.
- The repeated wording or other pattern, with a few comment links.
- Any prompt in the post that could explain the responses.
YouTube lets web users select Top comments or Newest first. Record which one you used. A ranked selection and a recent selection answer different questions. Neither makes the comments representative of everyone who watched.
Do not turn a sample proportion into an audience estimate. Fourteen repetitive comments out of 40 inspected comments does not mean 35% of the creator's followers are fake. You sampled comments, some people may have commented more than once, and you did not establish that repetition means automation.
Use evidence levels to choose the next step
These levels organize review work. They are not probability scores, platform classifications, or legal findings.
| Level | What you have | Next step | Safe record wording |
|---|---|---|---|
| Unavailable | Comments or relevant analytics cannot be inspected | Record the gap and seek a suitable alternative | Could not assess this field |
| Observed | A saved count, comment pattern, or discrepancy | Check selection method and content context | Pattern present in the inspected sample |
| Repeated | The pattern appears in several comparable posts | Ask a focused question and seek matching evidence | Repeated pattern needs explanation |
| Corroborated concern | Separate records support the same specific concern | Pause the booking for a second review | Evidence supports concern about the named activity |
| Decision recorded | Reviewer assessed the evidence and response | Proceed, limit the commitment, or decline | Decision based on the documented evidence gap or concern |
Two tools displaying the same underlying count are not independent corroboration. A screenshot of your own spreadsheet is not another source either. Ask what new fact each item adds.
If a provider labels followers suspicious, record that as its inference. Before using the estimate, read its published method, sample coverage, and update date. Keep unavailable details marked unknown. Do not convert a provider label into an observed fact about who paid for engagement.
Keep the claim narrow even when evidence is stronger. A record supporting activity on one video cannot establish that an entire audience is artificial. A booking decision also does not require a public statement about the creator's integrity.
Ask for context that can resolve the concern
Use a request tied to the observed material:
We saw repeated wording in the comments on the three linked videos during our review. Was there a viewer prompt or other activity around these posts? We would like to understand that context before choosing the campaign format.
If you need analytics, name the field, account, content, and period. Ask the creator to share only the relevant report, with unrelated information omitted. Do not request account credentials or private messages from followers.
For a YouTube review, Studio's Audience reports can add context, but the labels matter:
- Unique viewers is an estimate for the selected period. It does not certify that comments are genuine.
- Top geographies reflects watch time. It does not identify where every commenter lives.
- Some audience data can be limited. A missing field leaves a question unanswered.
Check timing before treating mismatched counts as evidence. YouTube says Analytics subscriber counts can lag public counts by about 48 hours while it performs verification and spam reviews. Compare like periods and capture dates first.
If the creator cannot share a relevant report, use the procedure for evaluating missing audience data. A lack of access changes the confidence of your decision; it does not establish deception.
Worked review worksheet
The following case and all its numbers are hypothetical. It illustrates a decision process, not a detection threshold.
| Worksheet field | Example entry |
|---|---|
| Decision | Whether to book a product demonstration video |
| Scope | Six recent demonstration videos; up to 20 newest top-level comments each |
| Observation | 18 of 120 inspected comments use one repeated phrase; 16 appear on one video |
| Initial evidence level | Observed |
| Alternative explanation | The video may invite viewers to repeat that phrase |
| Check | Watch that video's closing section and read its description |
| Hypothetical finding | The creator asks viewers to use the phrase when requesting a follow-up |
| Remaining uncertainty | The review cannot identify every commenter or the origin of every comment |
| Decision | Close this repetition flag; assess campaign fit using the other evidence |
Change the hypothetical finding and the next step changes. If there is no visible prompt and the repetition persists across several videos, move to a focused context request. If a material discrepancy remains unresolved after that request, pause or decline the booking on that basis.
Do not label the unresolved case proven fraud. Do not label the explained case a fully verified audience. Both conclusions exceed the review.
Assign one reviewer, a response deadline, and the evidence needed to reopen a closed concern. Then complete one worksheet for the specific pattern that could change your hiring decision.
