How the TrustGrowth Score works
One number from 0 to 100 for how much trust and growth a site has earned on the web, weighted 55 / 30 / 15 and never more than 25 points ahead of its standing.
Open our proof pageHeadline is at most Standing plus 25
- Headline, trustgrowth.ai
-
40of 100
Provisional, range 36 to 40.
- Standing
- 15+at least 15at least 40 − 25
- Slack
- 25never more
What is the score made of?
Three layers, weighted 55 / 30 / 15. Standing cannot be edited into existence; that is the thing being measured.
- 55%
Standing: Does the web treat this site as real?
The link graph (referring domains and domain rank, on the rank provider’s own curve), absolute search visibility, and corroboration: branded search demand and citations in AI answers. It is earned off-site and builds slowly.
- 30%
Integrity: Is the site well made?
Technical health, content (coverage, freshness, real search use, liveness), on-page E-E-A-T signals, performance and structure for generative engines. Fully in the owner’s control: a new site can honestly score 90+ on day one.
- 15%
Momentum: Is standing improving?
Measured over 28-day windows with hard volume floors. Below the floors it does not vote: its weight is shared out, never filled with a guess.
How far can polish carry a score?
25 points, no further. A site’s published score never runs more than 25 points ahead of what the web says about it.
Integrity may be high on day one; the headline may not. A flawless site with no standing reaches 25 at most: the cap stops polish passing for trust. Three companion rules bind as hard:
- Unknown is never a number. What we have not measured shows as provisional; missing evidence never becomes a mid-scale score.
- A measured zero is a zero. Zero referring domains, actually counted, scores zero, and the card tells it apart from “not collected yet.”
- Below-floor momentum does not vote. Growth on a handful of clicks is noise, so it is left out, not averaged in.
The hatched 25 is all the slack the contract gives above Standing, however well the site is made.
What does my number mean?
Read the band, not the instinct from school grades.
A well-built site under six months old usually lands in the 20s or 30s. That is correct, not a punishment: the web has not had time to gather evidence about it.
Early work moves Integrity, which is yours to fix today, and Momentum once search volumes clear the floors. Standing follows at the pace the web corroborates you, because that is what standing means.
| Score | What it means |
|---|---|
| Below 40 | Unknown or early. Most new sites live here, correctly. |
| 40–59 | Foundations in place; little external standing yet.trustgrowth.ai, 40 |
| 60–69 | Emerging. Tiers begin at 60. |
| 70–79 | Established: the first deserved 70s. |
| 80–89 | Strong. Uncommon. |
| 90+ | Rare. |
trustgrowth.ai · /proofs/trustgrowth · 21 Nov 2026
Why did every score drop in August 2026?
We replaced the scoring formula. The old one had three defects, each measurable in its own code.
72% of the weight measured how a site looks
Only 16% measured what it had earned. Technical polish, on-page signals and performance, all fixable in a week, ran the headline; links, traffic and mentions barely registered.
Missing evidence scored 50 out of 100
“We know nothing about this site’s backlinks” came out as a mid-scale number at full weight. Not knowing was worth more than measuring a zero.
The scales flattered the bottom
Full authority marks came at 500 referring domains, a nearly-new domain’s link rank mapped to 76/100, and the content score rose the younger the site was.
The result, on our own product
trustgrowth.ai scored 73, Silver tier, with zero organic clicks, zero top-100 keywords and seven referring domains. At the cutover the same site scored 32, provisional, with no tier.
A score that can put a brand-new site in the 70s measures craftsmanship, not trust. Ours did. At the cutover on 15 August 2026 every site on the platform dropped between 21 and 41 points.
We chose that on purpose: the alternative was keeping a number that meant nothing. The public production proof always carries the current number and re-syncs as evidence arrives.
Silver tier
provisional, no tier
- Organic clicks
- 0
- Top-100 keywords
- 0
- Referring domains
- 7
What do authority metrics give a new site?
Close to nothing. This is what a site under six months old scores on the industry’s established metrics.
That is not those products being harsh; that is what trust means. Google’s Search Quality Rater Guidelines tell raters to weigh what independent sources say about a site over what the site says about itself.
Lighthouse can give a brand-new site 100, because it scores the artifact and says so. Our old formula scored the artifact and called it trust. Now Integrity is the artifact layer, Standing the trust layer, and the headline is capped by Standing.
Re-baselining loudly has precedent: Moz’s Domain Authority 2.0 in 2019 retrained its model, dropped many scores overnight and published the method.
| Metric | Typical score under six months |
|---|---|
| Ahrefs Domain Rating | 0 |
| Moz Domain Authority | 1 |
| Semrush Authority Score | ~0 |
| Majestic Trust Flow | 0 |
| SimilarWeb / Tranco rankings | not listed |
What happens to my score history?
It is recomputed under the current formula, so your trend line stays comparable across the change.
-
History is re-expressed, not erased
Score history is recomputed under the current formula back to June 2026, wherever source data exists.
-
Recomputed history never ranks
Those points are stored as backfilled and are left out of every ranking.
-
The formula is versioned
Every published score carries its formula version, and every formula change goes through an audited activation trail.
-
Provisional means provisional
While the formula is provisional, or a score rests on incomplete evidence, the score card says so in words next to the number.
How is AI visibility measured?
As five separate readings, in sequence. They share no unit and do not compose, so nothing adds up to one score. A failed measurement is unknown, never a measured zero.
| Reading | What it reports | The rule | When it is unknown or withheld |
|---|---|---|---|
| 1. Reacha count of ten crawlers | GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, CCBot, Bytespider, Amazonbot and Meta-ExternalAgent each get a robots.txt permission check and a live homepage fetch under their own user agent. |
reachable_count = number of present crawlers, if none is unknown; otherwise null Only robots.txt HTTP 200 and 404 can establish permission. The llms.txt state is shown beside the count and never enters it. |
A crawler is unknown when the site or robots.txt is unreachable, or the fetch returns 403, 429, 500 or higher, or no status. Any unknown withholds the count; the known crawlers stay visible. A capture older than 35 days is stale. |
| 2. Readablea 0–100 score | An unweighted mean of four groups, each an unweighted mean of its checks’ pass rates over eligible pages: structured data (schema types); authorship and trust (author markup, publish date, update date); answer structure (heading hierarchy, FAQ section, FAQ markup, list usage); citation quality (outbound authority links). |
check_score = 100 × (1 − min(failures, eligible) / eligible) group_score = mean of defined check scores in the group Readable = round(mean of defined group scores) confidence = group_coverage × crawl_coverage It reads the newest complete crawl evidence, not Search Console. Readable is not the TrustGrowth Score’s GEO pillar. |
Noindex pages count in no denominator. A check with no eligible pages is omitted, not scored as zero, and so is a group with no defined checks. Confidence is coverage, not an interval: partial coverage lowers confidence, not the score. |
| 3. Retrieveda bounded census | A census of the keywords your domain ranks for, with no denominator, interval or sample size. Sampled AI Mode, ChatGPT and Gemini citation rates are a separate instrument: cited answers over answered rows per surface, each with a Wilson interval and its raw n, never combined with the census. |
cited_keyword_count = unique own-domain AI Overview keywords returned |
Each domain is capped at 200 rows, so a count at the cap is a floor, not a total. After 35 days the screen withholds the headline. |
| 4. Recalleda sample with a range | Each engine is asked the questions cold, with no web search: prompts × engines × samples. The headline is category questions. Branded questions are sampled and shown but kept out of it, because near-total branded recall is true of anything that exists. |
Every rate carries a 95% Wilson interval. Wilson, not the normal approximation, because samples run small and often give 0 of n, where the normal interval collapses to zero width and claims certainty. |
When the interval is wider than the rate, the percentage is withheld as “sample too small to call”; the count and its range stay. When two engines’ ranges overlap, the sample does not show that one knows you better. |
| 5. Impacta window, never a cause | The last 28 days against the prior 28, branded impressions only. |
Both windows ending three days back, because Search Console lands late and a partial final day reads as a decline that never happened. It is co-movement, never attribution. |
It prints nothing with no Search Console history, fewer than 56 days of history, no query containing your name, a zero prior window, or under 30 impressions. |
Questions we expect
Five we expect after a first score.
Why did my score drop?
The formula changed; your site did not. The old model overweighted what every site controls and scored missing evidence as 50/100. Removing both lowers almost every score, ours included. Nothing you built was un-built.
Is a 35 bad?
No. Below 40 is where most young or lightly established sites belong. A 35 with high Integrity describes a well-made site the web has not gathered much evidence about: most new sites, including ours at cutover.
Why does my site have no tier?
Tiers begin at 60 and mark earned external standing, so most sites do not launch with one. A missing tier is a stage, not a verdict.
Can I raise Standing quickly?
No, and that is the point. Standing is built from what the rest of the web does: linking to you, searching for you, citing you. Early work shows first in Integrity, then in Momentum once volumes clear the floors.
Where can I see this on a real site?
Our own snapshot is public and unedited, including the layers where we score poorly: /proofs/trustgrowth.
See every layer on our own site
Public and unedited, the weak layers included.