How AI visibility scores are calculated. Five dimensions, 100 points. Human-readable documentation for anyone who wants to understand or challenge the score.
The Sigbase Retrieval Score™ (SRS) measures how well a local business is positioned to be found, described, and recommended by AI answer engines. Sigbase measures three: ChatGPT, Gemini, and Perplexity.
The score does not measure website traffic, search rankings, or social media presence. It measures one thing: the signals that determine whether an AI platform recommends your business when someone asks.
Scores run from 0 to 100. Sigbase.AI publishes one number and one number only.
The SRS is composed of five measured dimensions across two tiers — Structural and Contextual.
Structural dimensions measure signals that exist on your website and can be directly improved. These four dimensions carry 73 of the 100 points.
Contextual dimensions measure observed citation behavior — what AI engines actually do when asked. Recommendability carries the remaining 27 points.
Each dimension is scored against an internal rubric, and the dimension results are normalized so that published weights sum to exactly 100. Weights below are the published 0–100 shares.
Entity Authority is not part of the score. Confirmation of a business by the wider web is real, but a substantial portion of it could not be measured directly — and Sigbase.AI does not assign points it cannot measure. Entity Authority was removed from the score in 2026 and exists only as a separate, optional add-on analysis outside the SRS. No guessed points appear in this score.
| Dimension | Tier | Points | Core Question |
|---|---|---|---|
| Findability | Structural | 13 | Can AI crawlers access and read your website? |
| Describability | Structural | 27 | Does your site give AI enough structured data to describe you accurately? |
| Summarizability | Structural | 20 | Can AI extract clean, citable answers from your content? |
| Comparability | Structural | 13 | Is your identity consistent everywhere AI might look? |
| Recommendability | Contextual | 27 | Do AI platforms actually recommend you when asked? |
| Total | 100 |
| Score | Grade | What It Means |
|---|---|---|
| 88–100 | A | Structurally dominant |
| 78–87 | B+ | Strong foundation |
| 70–77 | B | Established — gaps present |
| 55–69 | C+ | Vulnerable — action needed |
| 40–54 | C | Fragile — significant gaps |
| 25–39 | D | Critical — minimal foundation |
| 0–24 | F | Absent |
Can AI crawlers access your website? This dimension checks whether your robots.txt permits AI indexing, whether a sitemap exists and is reachable, whether your pages load within acceptable time, and whether your content is rendered in HTML rather than JavaScript. A business that AI cannot read cannot be recommended.
Does your site give AI structured data to work with? This is the highest-weight structural dimension because it is the most controllable high-impact signal. It evaluates the presence, completeness, and correctness of JSON-LD schema markup — specifically LocalBusiness, Service, and FAQPage schema types — as well as Open Graph metadata and NAP (name, address, phone) consistency. AI platforms rely heavily on structured data to accurately describe businesses in generated responses.
Can AI extract clean answers from your content? This dimension evaluates whether your pages contain enough structured, substantive content for AI to summarize you accurately. It checks heading hierarchy, FAQ section presence, service section depth, and whether your content answers the five core questions a prospective customer would ask: what the service is, who it is for, what the process involves, what outcomes to expect, and what it costs.
Is your identity consistent across every surface AI might use? Name, address, phone number, and category must match across your website, schema markup, Google Business Profile, and major directories. AI platforms stitch together information from multiple sources — inconsistency creates conflicting signals that reduce citation confidence.
Do AI platforms actually recommend you when someone asks? This is the only dimension that measures observed behavior rather than inferred readiness. Sigbase runs structured retrieval queries — brand queries, category queries, and comparison queries — against live AI search systems and records whether your business appears in the results. A business can score well on every structural dimension and still not appear when someone asks. Recommendability captures that gap directly.
Each business is scanned by the Sigbase measurement instrument — an automated system that performs the following in sequence:
A dimension that cannot be measured is recorded as null, never as zero. Zero is a claim: the engines were asked and did not name the business. Null is the record that the engines were not reached. A scan carrying an unmeasured dimension is excluded from banding — it publishes no total, no grade, and no band — and the remaining dimensions are never rescaled to stand in for the missing one, because a four-dimension total presented on a five-dimension scale would place two different instruments on one grade and silently break every comparison in the index.
Every measurement is anchored to a declared market rather than to a business's mailing address, because a business competes in the market its customers search and that is not always the city on its letterhead. The anchor market is recorded with each observation and stated on the report. Businesses are compared and ranked only against others sharing the same anchor. A business's anchor is frozen across readings; changing it begins a new dated series rather than continuing an existing one.
Sigbase.AI re-measures tracked businesses on a recurring cycle. Because measurement runs under a fixed daily volume limit, a full cycle across the tracked population spans a window of several days rather than a single moment — published figures state that window. Score history is preserved, and every score is labeled with its measurement date and method version.
| Version | Date | What Changed |
|---|---|---|
| Method 2.2 | August 12, 2026 | A dimension that cannot be measured is recorded as null, never as zero, and is excluded from banding: a scan carrying an unmeasured dimension publishes no total, no grade, and no band. Zero is a claim — the engines were asked and did not name the business. Null is the record that the engines were not reached. Because a scan that would previously have published a reduced total now publishes none, totals are not comparable across this boundary. Two documentation corrections ship in the same revision with no change in behavior: the measured engine set is stated as ChatGPT, Gemini, and Perplexity, and anchor-market handling is described explicitly. |
| Method 2.1 | August 10, 2026 | Adds a direct identity check alongside the scoring queries, so a zero can be distinguished from non-recognition. Evidence only — the scoring formula is unchanged and scores remain comparable across this version. |
| Method 2 | July 2026 | Entity Authority removed from the score: unmeasurable signals are never assigned guessed points. Five measured dimensions, normalized to 100. Three answer engines, equally weighted, two-of-three consensus. Grade bands frozen July 19, 2026. |
| SRS v2.0 | April 9, 2026 | 100-point scale replacing the 50-point scale. Current scale, still in force. |
| SRS v1.0 | March 2, 2026 | Initial scoring methodology. 50-point scale, five dimensions. Retired. Entities scored under v1 are not displayed in rankings or city pages until rescanned under v2. |
Sigbase Retrieval Score™ is a trademark of Sigbase.AI. SRS v2.0. Method version 2.2. Grade bands frozen July 19, 2026.