01
Freeze eligibility before scoring
A record must have a product-scoped, complete and comparable Semrush domain row. A failed gate receives no score or rank.
AI Search Evidence Proxy v0.1
A narrow score with a visible method.
This methodology compares the domain-level search-evidence footprint surrounding 100 AI products. It shows how 78 comparable records are scored, why 22 scores are withheld, and which claims the index does not make.
Method at a glance
Freeze. Normalise. Weight. Rank.
The method answers one question: how large is each eligible record's measured search-evidence footprint relative to the other eligible records in this dated cohort?
02
Reduce extreme scale effects
Each non-negative input is transformed with ln(1+x), then min-max normalised from 0 to 100 inside the 78-record eligible cohort.
03
Apply the fixed seven-input formula
The normalised inputs are weighted exactly as published below. No hidden adjustment or source-list position enters the score.
04
Rank the retained value
Six decimal places determine rank. The public score is rounded to one decimal, with an explicit tie rule and 78-record denominator.
Normalisation and weights
Every scored input is visible.
All required inputs are non-negative. If an eligible-cohort metric had no range, that component would receive a deterministic value of 50. No metric is degenerate in this snapshot.
Per-metric normalisation
100 x (ln(1 + value) - minimum eligible ln(1 + value)) / (maximum eligible ln(1 + value) - minimum eligible ln(1 + value))Google AI Overview keywords
40%The measured AI Overview keyword footprint for the associated domain.
Commercial-intent positions
15%Commercial-position evidence in the Semrush US snapshot.
Commercial-intent traffic
10%Estimated traffic attached to the measured commercial positions.
Organic keywords
10%The domain's measured organic keyword footprint.
Organic traffic
5%Estimated organic traffic in the dated database snapshot.
Authority score
10%The provider's domain authority signal, normalised within the cohort.
Referring domains
10%Measured referring-domain evidence for the associated domain.
Pillar summary
40 / 40 / 20AI Overview footprint / commercial plus organic evidence / authority plus referring domains.
Eligibility first
Missing evidence is not scored as zero.
A record receives neither score nor rank when its domain evidence cannot support a complete, product-to-product comparison. The record remains visible in the directory with the reason attached.
Product scope needs review
The listed entity is too broad, uncertain or insufficiently product-specific for the comparison.
Metric scope mismatch
The available domain evidence describes a broader parent site rather than the requested product.
Required evidence unavailable
A complete Semrush row or required formula input is unavailable.
Duplicate canonical domain
Multiple products share one measured domain, so the same domain signals cannot support separate product ranks.
Reason counts can overlap. Together they affect 22 unique records; they should not be added as though they represented 26 different products.
Calculation rules
Rounding never decides rank.
Retain six decimals
The underlying weighted score keeps six decimal places for ranking. The directory rounds only the displayed score to one decimal.
rank value: 6 dpShare competition rank
Records with identical six-decimal scores receive the same competition rank. Candidate ID only stabilises display order and never breaks a tie.
1, 2, 2, 4Avoid double-counting
Semrush rank, backlinks, referring IPs, follow or nofollow links, and source-publisher position are excluded from the formula.
evidence retained, weight 0%Responsible interpretation
What the proxy cannot tell you.
The result is useful only when the scope stays narrow. A larger or more established domain can score strongly because this version measures domain-level evidence, not the product experience or every AI answer surface.
01
No direct assistant observations. This version does not query ChatGPT, Claude, Perplexity, Gemini, Copilot or another assistant to measure product mentions, positions or citations.
02
No product-quality judgement. A score does not measure usefulness, accuracy, customer satisfaction, adoption, market share, safety or value.
03
No universal rank. Results compare only the 78 eligible records inside this 100-product cohort and this dated database snapshot.
04
No proof of recommendation or citation. Search evidence can support a future research question, but it is not evidence that an assistant selected or cited the product.
05
No source-list relabelling. G2 or Product Hunt order remains discovery-source evidence and is never presented as a Brandilite rank.
Read the score beside its evidence receipt.
The directory shows every product, the eligible denominator, each available input, the strongest measured pillar and an explicit reason whenever scoring is withheld.