SpyderBot Documentation
Track how AI sentiment toward your brand changes
AI sentiment is improving overall, but your team still sees unfavorable answers. Or a negative result appears even though product information is accurate. What should you respond to?
This playbook helps you read the current sentiment, its change over time, and the underlying claims. The goal is to identify a portrayal issue worth addressing, not to make every answer sound positive.
Your outcome: a specific sentiment or portrayal pattern, its factual interpretation, and a decision to correct information, clarify positioning, investigate, or monitor.
1. Start with the right sentiment scope
Open the benchmark's Brand Understanding observations and the Controlled AI Sentiment result. Confirm the brand, Context, AI systems, and actual observation dates. For a competitor comparison, use the controlled comparison scope provided by the product.
These observations concern how AI evaluates the brand under the benchmark's controlled brand contexts. General AI Brand Portrayal in other observations can provide useful investigation evidence, but it is a different aggregate scope. Do not join the two into one trend line.
If you only have a current observation, assess its claims now and establish future comparison. Missing sentiment evidence is not a neutral or zero result.
2. Read the mix as well as the summary
Review the current sentiment and What Changed?, then inspect AI Sentiment Over Time and Sentiment Mix Over Time. Confirm the compared dates and the entity selected in the mix.
Positive, Neutral, Mixed, and Negative describe different portrayals. A mixed answer may combine a real strength with a relevant limitation. It should not automatically become a correction task.
| Observed pattern | What to inspect next |
|---|---|
| Positive portrayal rises while negative portrayal also rises | Look for different requirements or systems moving in different directions. |
| Negative portrayal falls while neutral portrayal rises | Check whether criticism disappeared or answers became less informative. |
| Mixed portrayal increases | Inspect the trade-offs being stated and whether they are accurate. |
| The aggregate is stable but one important claim worsens | Assess the claim's business significance rather than dismissing it because the headline is unchanged. |
| A competitor is portrayed more positively | Compare the relevant controlled contexts and underlying attributes, not only the scores. |
Read the mix alongside its eligible evidence and coverage. If fewer contexts or systems were successfully evaluated, a higher positive share may describe a changed evidence set. Check whether the comparison is supported before treating it as improved portrayal. Unknown, unclassified, or missing evidence should not be interpreted as neutral. This is a reason to inspect the product's coverage information, not to recalculate the score yourself.
A favorable aggregate does not make every claim accurate. An unfavorable aggregate does not establish that every criticism is false. Choose the claim that affects a customer decision, using the evidence available for that claim.
3. Find the statement behind the movement
Start from the system's relevant change or portrayal finding. Open the affected contexts through the available detail view or Observatory. Follow the summary to the response evidence needed to understand the claim; do not manually classify every answer.
Read the qualification around the statement. "Suitable for small teams, but limited for complex enterprise approvals" contains a strength and a limitation. Determine whether the answer is relevant to the intended audience before responding.
When comparing dates, check whether the same issue became more prominent, a new issue appeared, or coverage changed. If a driver remains unresolved, keep it unresolved. A new cited source is an investigation lead, not proof that it caused the sentiment change.
4. Choose the response that improves accuracy and customer understanding
| Verified interpretation | Useful action |
|---|---|
| A factual claim is incorrect or outdated | Correct inconsistent owned information or prepare an evidence-backed external correction. |
| A limitation is accurate | Clarify fit and trade-offs; involve Product if the limitation matters to the target audience. |
| A genuine strength is missing in relevant answers | Improve the specific public evidence that helps customers evaluate that strength. |
| AI confuses entities or product ranges | Investigate identity and entity representation. |
| A subjective judgment has no clearly verified factual error | Inspect its reasons and monitor the relevant pattern; do not label it false solely because it is unfavorable. |
| The response and extracted sentiment disagree | Preserve the example for support before acting on the interpretation. |
Choose one claim or attribute with meaningful customer impact. A narrow verified correction can be useful even if a broad trend is not yet established. Larger repositioning work needs evidence appropriate to that investment.
Example: positive and negative portrayals grow together
Illustrative example. A retailer sees more positive portrayals of product quality, but also more negative statements about returns. Reading only the positive share would miss a problem affecting purchase confidence.
The team inspects the returns statements. Some accurately describe a restriction for customized items; others incorrectly apply that restriction to all products. It preserves the true restriction, clarifies the standard returns information, and checks related source material.
The task is to improve the accuracy of returns portrayal. The team evaluates that claim separately from the overall sentiment result.
5. Set a targeted review
After implementation, verify the changed information. In later comparable observations, check the specific claim and its scope first, then review the broader mix. If the claim is accurate but mixed sentiment remains, assess whether the remaining trade-off is useful and truthful.
If sentiment improves while visibility declines, report both observations separately. You may be portrayed more favorably in controlled questions while being introduced less often in discovery. Neither result cancels the other.
You are done when you can name the relevant sentiment pattern, explain the underlying claim, and justify the response. Use Improve how AI describes and positions your brand for the detailed correction work, or Investigate a change in your AI visibility for a broader change investigation.