ContextHint
Measurement guide

Which Context Hint Is Working?

Keep the theme associated with better qualified business outcomes—after confirming both tests had comparable eligibility and enough opportunity.

The decision rule

Business quality beats delivery volume.

Public captured-ad evidence can help form a hypothesis. Only your own campaign and conversion data can tell you whether that hypothesis produced valuable outcomes.

Evidence ladder

Know what each signal can prove.

01

Observed market evidence

Captured ads, advertiser presence, and prompts where ads appeared can show that a theme exists in the market. They do not reveal an advertiser's literal private hint or conversion rate.

02

Inferred targeting evidence

Repeated creative and prompt patterns can support a reconstructed context-hint hypothesis. Confidence describes the reconstruction, not campaign success.

03

Your delivery data

Impressions and reach show that the complete campaign entered relevant opportunities. They do not isolate the causal effect of one sentence.

04

Your business outcomes

Qualified conversions, revenue, retention, or another downstream metric are the strongest decision signals available to the advertiser.

Worked interpretation

Fewer impressions can still produce the better theme.

Suppose a comparison-intent group receives less delivery than a broad research group but produces more qualified trials and retained customers. The broad theme wins on reach; the comparison theme wins for the business.

  1. 1

    Confirm both themes were eligible.

  2. 2

    Check whether each reached the intended buyer.

  3. 3

    Compare qualified downstream outcomes.

  4. 4

    Record other variables that also changed.

Choose on the complete funnel, not one convenient metric.

Designing the comparison first? Use the testing workflow.

Evidence boundary. ContextHint can analyze captured ads, prompt patterns, inferred themes, and similar advertisers. It cannot see another advertiser’s private settings, spend, auction history, or conversions.