What a DecodeIQ result measures.
Six frozen questions, three engines, up to 54 recorded answers. Here is how each number is produced.
One product. The same six questions.
Drafted from product identity and category, then frozen.
The same six questions run across three engines in three waves.
They do not come from a Voice Map.
Illustrative product and sample questions.
- What are the best lightweight backpacks for day trips?
- Which backpacks fit under an airplane seat?
- What backpacks are comfortable for a day of sightseeing?
- Which everyday backpacks should I consider for under $100?
- Which backpacks keep a laptop and water bottle in separate pockets?
- Which daypacks have waterproof fabric for rainy hikes?
Six questions × 3 engines × 3 waves = 54 answers
| Engine | Wave 1 | Wave 2 | Wave 3 |
|---|---|---|---|
| ChatGPT | 6 | 6 | 6 |
| Gemini | 6 | 6 | 6 |
| Perplexity | 6 | 6 | 6 |
Page Readiness is page analysis.
40% Technical Readiness + 60% Buyer-Language Fit, rounded.
This is not the paid product-appearance metric.
The live ChatGPT result is a separate observation.
Technical Readiness
40%
- Pass
- 100
- Partial
- 50
- Fail
- 0
Non-applicable checks are excluded.
Buyer-Language Fit
60%
- Buying criteria10 / 5 / 0
- Objections10 / 5 / 0
- Use cases10 / 5 / 0
- Outcomes10 / 5 / 0
- Comparison anchors10 / 5 / 0
Buyer phrasing 10 · Specification phrasing 5 · Absent 0
Seller-language penalty
- None
- +10
- One or two
- +5
- Three or more
- +0
Total / 60 × 100, rounded
If no technical checks apply, the score uses Buyer-Language Fit alone.
Read engine status before comparing.
Collection methods differ by engine.
A check is charged only when at least one engine completes all six questions across all three waves.
Detection statistics require the full 54-record set.
| Engine | Collection method | Grounding | Completed answers |
|---|---|---|---|
| ChatGPT | Third-party collection endpoint with snapshot fallback. | Provider web-search flag when supplied; otherwise unknown | 6 × 3 = 18 |
| Gemini | See definitions | Unknown, excluded from grounded comparisons | 6 × 3 = 18 |
| Perplexity | Live LLM response endpoint. | Recorded when citations or search queries exist; otherwise unknown | 6 × 3 = 18 |
Same questions. See what changed.
Compare the same stored questions.
Scorer and normalizer versions must be compatible.
A changed question set is not a valid comparison.
| Recorded field | Check 1 | Check 2 |
|---|---|---|
| Product appearances | 0/54 | 39/54 |
The change followed an update to DecodeIQ's product-matching rule. Product content did not change between checks.
Observed change does not establish cause.
We do not claim that generated content moves visibility scores.
A re-check shows what changed in the recorded answers. It does not show that your content caused the change. A change also does not establish traffic, sales or revenue from the recorded answer. Review content as an edit to evaluate, then inspect later evidence.
Do not treat a before-and-after example as proof of uplift.
Full definitions and limits
What a DecodeIQ result measures
One product. The same six questions.
No prompt slots to fill.
Enter a product URL. DecodeIQ drafts six unbranded questions from the product and category, then freezes them for later checks. Each question runs across ChatGPT, Gemini and Perplexity in three waves, for up to 54 answers. The question text and product identity stay frozen for re-checks.
Questions come from product identity and category. They do not come from a Voice Map.
Page Readiness is page analysis.
The live ChatGPT result is a separate observation.
Page Readiness is rounded from 40% Technical Readiness and 60% Buyer-Language Fit. Technical checks score pass as 100, partial as 50 and fail as 0. Non-applicable checks leave the average. If none apply, the score uses Buyer-Language Fit alone.
This is not the paid product-appearance metric.
Buyer-Language Fit reads five page signals: buying criteria, objections, use cases, outcomes and comparison anchors. Each receives 10 points for buyer phrasing, 5 for specification phrasing and 0 when absent. No seller-language examples adds 10 points; one or two adds 5; three or more adds 0. The page-analysis total is divided by 60, multiplied by 100 and rounded. Live mentions and winner-language alignment add no points to the current score. A language classifier is involved, so this is a diagnostic reading, not a measured share of shoppers.
Older v1 cached rows can contain the prior blended calculation. Compare results only within a compatible report version. The readiness bands describe page checks, not the chance of a recommendation.
Keep collection limits beside the result
Read engine status before comparing checks.
A paid partial check is chargeable only when an engine completes all six questions across all three waves. A check with no completed engine is not charged. Detection statistics require the full 54-record set. A partial result can show evidence without those statistics.
Gemini grounding is unknown and excluded from grounded comparisons.
Collection methods differ by engine. The current paid Perplexity collector uses a live LLM response endpoint under the shared record contract. Do not read engine names as a promise of identical consumer-interface collection. Answer context, model changes and response variation can affect a comparison. The current product does not support a claim that every shopper saw the recorded answer.
Compare the same stored questions with compatible scorer and normalizer versions. A changed question set is not a valid same-question comparison. An accessible page can be recorded as an intervention surface. Access may be unknown or blocked. For an Amazon-listing-only edit, keep external-crawler observations separate from Amazon-owned shopping systems.
Observed change does not establish cause.
We do not claim that generated content moves visibility scores.
A re-check shows what changed in the recorded answers. It does not show that your content caused the change. A change also does not establish traffic, sales or revenue from the recorded answer. Review content as an edit to evaluate, then inspect later evidence.
Do not treat a before-and-after example as proof of uplift.
Is AI naming your product
or someone else’s?
Start with your product URL. See what comes back.