We Cut Our Own Trust Score From 73 to 32

On 15 August 2026 we replaced the formula behind the TrustGrowth Score. Our own site's number fell from 73 to 32, and all six scored sites in our fleet dropped between 21 and 41 points on the same day. The old formula, the new one, and the full weighting are documented at our methodology page.

Start with the embarrassing part. trustgrowth.ai — the website of a product whose entire job is scoring website trust — was five months old and scored 73 out of 100, Silver tier. On the day we measured it, that site had zero organic clicks, zero keywords ranking in the top 100, and seven referring domains.

Nothing was broken. The formula was doing exactly what it had been designed to do. That was the problem.

How a site with no traffic scores in the seventies

Sort any compound site score into two piles. How the site looks — technical health, page speed, markup, on-page signals — is everything an owner can fix in a week. What the site has earned — links, traffic, mentions — is everything only the rest of the web can grant.

In our old formula, 72% of the weight sat on the first pile and 16% on the second.

That alone guarantees inflation: any competently built site starts with most of the score already in hand. Two quieter mechanisms did the rest.

First, missing evidence was scored as 50 out of 100. If we had no backlink data for a site, the authority component didn't sit out; it contributed a 50, at full weight. "We know nothing" outscored "we measured zero." Not knowing was worth points.

Second, the scales flattered the bottom. Full marks for referring domains arrived at just 500 of them, and a nearly-new domain's link rank mapped to 76 out of 100. Our content score was actually higher for younger sites: a freshness calculation that read "no old content" as "all fresh content."

Stack those together and a polished young site doesn't sneak into the 70s. It belongs there. Our site cleared 70 on the strength of the first defect, was held up by the second wherever evidence was missing, and was rewarded by the third for the very thing that should have counted against it: being new. The formula was measuring craftsmanship and calling it trust.

The industry already agreed on what a site like that scores

For a domain under six months old with no links and no traffic:

Metric Typical score What it measures Ahrefs Domain Rating 0 Backlink profile strength Moz Domain Authority 1 Predicted ranking ability from link data Semrush Authority Score ~0 Link power, traffic and spam signals Majestic Trust Flow 0 Trust propagated from vetted seed sites SimilarWeb / Tranco not listed Traffic and popularity ranking

Zero, one, zero, zero, absent. This table is methodology grounding, not a ranking of these tools or an endorsement of any of them. They disagree about almost everything else — scale, inputs, philosophy — but they agree completely here, because they all define trust the same way: external corroboration. Google's Search Quality Rater Guidelines direct raters to weigh what independent sources say about a site's reputation over what the site says about itself. Trust you can manufacture by editing your own pages is not trust; it's testimony.

The honest exception proves the rule. Lighthouse will give a brand-new site a 100, because Lighthouse scores the artifact, and says so on the label. The dishonest position is the one we were in: scoring the artifact and calling it trust.

The redesign, in three sentences

The score is now built standing-first. Standing carries 55%, Integrity 30%, and Momentum 15% — Standing being what the web says about you through links, search footprint, branded demand and citations in AI answers; Integrity being how well the site is made; Momentum being whether Standing is improving. One published contract binds the whole thing: the headline can never exceed Standing + 25, so no amount of polish puts an unknown site above the low 30s.

Three companion rules do the rest of the work:

• Unknown is never a number. Missing evidence shows as provisional instead of quietly becoming a mid-scale score.
• A measured zero is a zero. Zero referring domains, actually counted, scores zero, and the card distinguishes that from "not collected yet."
• Below-floor momentum does not vote. Growth computed on a handful of clicks is noise, so it is omitted rather than averaged in.

The receipts

Here is what the cutover did to our own fleet. Sample: all six scored properties we operate, n=6, both columns measured 15 August 2026 on the same day's run. Each one has a public proof page.

Site Old model Standing-first Change trustgrowth.ai 73 (Silver) 32 −41 metrixnest.com 57 19 −38 techwright.io 52 23 −29 gleania.com 50 25 −25 nferens.com 49 26 −23 devdiary.me 48 27 −21

One number to carry away, so it isn't split across two sections: the cutover value for trustgrowth.ai is 32, measured 2026-08-15, and the current value is whatever its proof page says today, which was 35 on 2026-08-21. The first is history; the second is the live number.

Look at what happened to the order, not just the values. Under the old formula metrixnest.com ranked second in that fleet and devdiary.me ranked last. On the cutover run the order was exactly reversed: devdiary.me second, metrixnest.com last. Nothing about either site changed overnight; what changed is which signal the formula listened to.

You can still see the gap that was being drowned out. Authority is the pillar fed by the link graph, and it is public on every proof page: on 21 August 2026 devdiary.me's Authority reads 22 against metrixnest.com's 9. The old formula had been ranking the two by polish, and polish was loud enough to bury a difference that size. A ranking that inverts under an honest measurement was never a ranking.

What this measurement does and doesn't show

Three limits worth stating plainly, because a before/after table invites more confidence than it has earned.

These are dated values, not current ones. Scores re-sync as evidence arrives. Our own site read 35 on 21 August 2026, six days after the 32 in the table above. The number moved without the formula changing again. Every proof page linked here shows today's figure, which is the one to trust; the table is a snapshot of one day.

History is re-expressed, not erased. Score history is recomputed under the current formula back to June 2026, which is the only way a trend line stays comparable across a formula change. It also means the 73 no longer appears on our public trajectory chart — that chart now shows what our site would have scored under today's rules. If you go looking for the old number on the proof page, it is deliberately not there.

Six sites is not a study. This is a descriptive before/after on properties we happen to operate, not a sample designed to represent the web. They are also all ours, which is a selection bias worth naming out loud: they share hosting, build practices and a technical baseline that most sites do not, so their Integrity layer starts unusually strong. The table shows the direction and rough size of the change on real sites. It does not establish what your site will do.

Reading the new numbers

Under the new bands, below 40 is "unknown or early", which is where most young sites belong, ours included. Tiers don't start until 60, and 70–79 now reads "established". We mean that label: those are the first 70s the formula can no longer hand out for free.

Our Integrity layer still scores high, and that is the design working, not residual flattery. A new site should be able to prove it is well made from day one. It just doesn't get to call that trust any more.

We are not the first to do this in public. Moz rebuilt Domain Authority in 2019, retrained the model, dropped a great many scores overnight, and published the reasoning rather than letting people discover it from their dashboards. That is the pattern we followed here.

What this means if you're scored

If your site is new: expect the 20s–30s, with Integrity doing most of the talking. That is not a punishment; it is an accurate statement that the web hasn't accumulated evidence about you yet. The work you do now shows up first in Integrity, then in Momentum once your search volumes clear the noise floors. Standing arrives at the speed the web corroborates you, which is exactly why it is worth having a number for.

If your site is established: your score dropped too, and your relative position almost certainly improved — because the sites that used to score within a few points of you on polish alone just lost that subsidy. When you see a 70 now, it was earned.

If you're comparing tools: ask any score you rely on two questions. What does it give a site nobody links to, nobody searches for, and nobody cites? And can a number that answers "73" to that mean anything when it says 73 about someone else?

A low, honest number you can move beats a high number that means nothing. That is the whole redesign in one line.

Questions we expect

What does the new formula do to a brand-new site's score?
It puts it in the 20s–30s, with Integrity carrying most of the weight, because Standing has not accumulated yet.

Why did the ranking between two of your own sites invert?
The old formula ranked them by polish; the new one ranks them by what the web says about them, and the gap in their link graphs was much wider than the gap in their build quality.

Is my old score gone?
No. History is recomputed under the current formula back to 2026-06 so the trend line stays comparable, but the old number itself is no longer displayed.

Can I raise Standing quickly?
No. Standing is built from what other sites and searchers do, which is deliberately outside your control. Early work shows up in Integrity first.

Does a low score mean my site is bad?
No. Below 40 is "unknown or early", which describes most young sites accurately. A missing tier is a stage, not a verdict.

Run it on your own site

The full methodology, layer weights, the cap, the bands, and what we can and cannot measure, is public at /methodology. See it run against a real site, unedited, at /proofs/trustgrowth. If you want the same unedited answer about your own site, run a free scan.