Evidence behind the findings: examples, conflicts and limits

In an AI check

Open Evidence behind the findings in a new check. It shows exact claim excerpts from passages across the saved answers, the assistants behind them, available source links and page dates. Repeated wording is counted separately from assistant variety. A cited page is not proof that a claim is true, and limited support is labelled. The same evidence summary appears in the PDF, PowerPoint and Markdown exports.

New checks select relevant passages from across the saved answers and share coverage across assistants. The claim list shows consequential statements with exact supporting excerpts, rather than treating an assistant's opinion as a fact. Open a finding to see its examples and available source links. Older checks may lack these fields; the upgrade cannot recover text they did not retain.

Possible conflicting claims

It flags possible conflicts when matching statements about the same question have different values or negation. For a few unresolved price, policy or product claims, it can review up to 3 already-discovered pages on the supplied official website. You see a related passage, an unavailable page or no match. Review dates and conditions before deciding what is correct; PollyX does not assign a truth verdict.

The comparison is conservative: matching wording, question and category with changed values or negation. Different explicit dates or conditions are kept apart. Differently worded disagreements can be missed. The absence of a warning does not establish agreement.

Primary-source review reads at most 3 already-discovered official-site pages when suitable pages are available. It respects page access rules and has a bounded page-reading allowance. An unavailable page or no matching passage leaves the claim unresolved. Your Correct, Wrong or Outdated verdict remains yours to set.

In a consumer report

Evidence behind the findings shows the source records linked to each theme, host variety, known dates and known copies of the same story grouped together. Few linked sources mean limited evidence; different hosts do not prove independence. When searches repeat stories or miss the subject, new reports can redirect the remaining search allowance toward missing themes, source types or markets within the requested scope. The total allowance stays fixed, and searches never aim for a particular positive or negative balance.

The allowance is up to 12 web searches in total across resumed analysis calls. Coverage can improve when an unproductive query is redirected, but it is not exhaustive. Source hosts, dates and known copied stories describe the linked records; they do not establish independent people, a representative sample or a statistical confidence level.

AI next to consumer evidence

Yes, when you also have a compatible consumer sentiment report. The reports must match the brand, market and language, cover consumers, and be within the preceding 30 days at the time of the AI check. Linked examples use exact excerpts about the same brand. Opposing views require the same proposition and scope with explicit negation; different surveys or ratings do not qualify. Concerns may be absent from the sampled passages without being absent from all AI answers. Older comparisons without context-checked excerpts hide their theme interpretations. Missing coverage does not prove a claim false. The consumer report uses its own check or credit, and the two products' scores are not compared as if they were the same measure.

New linked examples appear when the comparison is generated with both compatible results available. Existing comparisons are not automatically regenerated. Older comparisons without context-checked excerpts hide their theme interpretations while retaining the matched report and separately defined scores. Each new comparison keeps exact substantive excerpts about the right brand; headings and separators are not evidence. Different metrics or survey populations cannot establish opposing views. Original citation pages are retained per answer on new checks. Older checks may retain citation hosts without page links; the card distinguishes these. Consumer theme references must belong to the saved source list, while quote links require exact text matches. Some review evidence links to the store or outlet page that was collected because an individual review link is unavailable. A concern missing from the selected passages is labelled as missing from that sample, not from every AI answer.

What the comparison chart measures

The AI visibility tools comparison counts six equally weighted documented research areas in each displayed plan. PollyX already covers all six. Better passage selection, evidence inspection and limited primary-page review deepen those areas without adding artificial points. The chart measures coverage, not accuracy, research quality or proof that a claim is true. Plan limits and tracking frequency differ; unverified features and paid upgrades are not counted.

Reading and sharing the card

Each card explains the data it shows, and each finding explains its counts or comparison basis. The PDF, PowerPoint and Markdown carry the same evidence summary. Some source text cannot be redistributed, so exports retain permitted links and titles rather than copying restricted material. See How quotes are checked.