How we estimate the impact of fixes

We have not measured what your traffic would do. We distribute a fixed, published budget of potential across the criteria that apply to your site, and weight each by how broken it currently is. Two sites with the same failures get the same numbers. Nothing here is a promise about outcomes.

The score answers one question: how ready the site is for AI systems to understand and recommend it. So the criteria are split into three classes, and they weigh very differently.

CORE-AI — 54 criteria, directly about AI adaptation: Schema.org and JSON-LD, AI summaries, answer structure, FAQ quality, machine-readable identity, AI-crawler access. Weight ×6. SUPPORT — 50 criteria: clear services, evidence behind claims, a consistent name across the web. Weight ×1.2. HYGIENE — 24 criteria of ordinary site hygiene: HTTPS, pages that answer, legal pages. Weight ×0.05.

Points are earned for AI preparation done; the basics are required but add nothing. 23 criteria are marked passive — any site passes them without doing anything: not blocking a crawler, serving HTML, not putting noindex on its own pages. A passive criterion never enters the average — neither side of it, whether it passed or failed. The report shows it as it is: a pass as a pass, a failure as a problem to fix. Blocking an AI crawler is a defect and the report names it; it is not measured by the score, because not blocking one is no achievement.

The overall score is the average of every counted criterion, weighted as above. A category’s weight in the report is the sum of what its counted criteria carry, which is why the “how the score is calculated” breakdown adds up to the total by construction.

The report panel shows the rise as a percentage of your current base in the channel. The channel score is your category scores weighted by the channel shares (table below). The model computes a multiple of the base and the report states it as a rise: a multiple of 1.5 is “a potential gain of: +50%”.

strength(балл) = (балл / 100) ^ γ            γ(SEO)=1.0, γ(AI)=2.0
база           = strength( max(балл_канала, 25) )
достижимое     = strength( балл при всех исправленных Failed/Partial )
кратность      = достижимое / база
прирост        = (кратность − 1) × 100 %      → «до +N %»

Channel strength = (score / 100) to the power γ. For SEO γ=1 (linear); for AI γ=2 (squared): the same score lift on a low base yields a much larger relative gain — the AI channel is younger and growing faster.

The base is taken from a score of no less than 25: even a weak site has a non-zero base — brand, direct visits — so the rise is finite. The same floor caps the maximum: for AI the ceiling is (100/25)² = a multiple of 16, that is “a potential gain of: +1500%”.

When the AI-channel readiness score is below 40, the report panel shows the fact and two numbers — current readiness and achievable readiness — alongside the rise. The threshold is not a round number; it is read off the actual distribution. Across the 126 completed audits carrying a v2 base: below 40 the rise runs +610% to +1500%, with six audits sitting exactly on the model’s ceiling; from 40 to 50 it runs +310% to +510%; from 50 up, +50% to +290%.

A rise of “+1400%” on a base of 26 is arithmetically correct and unreadable: it means multiplying almost nothing, and its size is set by the base floor of 25 rather than by the site — when a site scores exactly the ceiling, the model is describing its own floor, not the site. So below a base of 40 we lead with the fact — AI systems currently bring you almost no visitors — show readiness as “now → achievable” out of 100, and keep the percentage as the secondary figure. The 40 threshold is 1.6× the model’s floor; 14% of audits fall under it.

The calculation is identical either way: the model computes the same channel strengths and the same multiple; only which of its numbers leads changes. A site with a live AI channel (base at or above the threshold) sees the multiples described above.

A budget is the total upside the model will ever attribute to one channel, for a maximally broken site. Fixing everything cannot yield more than this, because the model never distributes more than it has.

The AI budget is roughly three times the SEO one. That is a modelling assumption, not a finding: AI-sourced traffic starts from a low base and is growing, so adapting for it yields a larger relative gain than further tuning of a mature search channel. These are calibration constants — revising them means issuing a new model version.

Each channel budget is divided between the 11 audit categories by fixed shares. Each column adds up to 100%.

Within a category, a criterion takes a portion proportional to its weight in our methodology — the same weight that drives the score. We do not keep a second set of weights for impact, because two weightings would eventually disagree.

liveShare(K, ch) = SHARE(K, ch) / Σ SHARE(K', ch)
potential(c, ch) = BUDGET(ch) × liveShare(K, ch) × Weight(c) / Σ Weight(K)
impact(c, ch)    = potential(c, ch) × statusFactor(c)

The normalizer matters when a whole category does not apply to your site. Its share is redistributed across the categories that do, so the budget stays the ceiling for a maximally broken site of your kind — rather than making some kinds of site structurally capable of less.

• Failed — the full potential is still on the table.
• Partial — half of it.
• Passed — nothing left to gain here.
• No verdict at all (not applicable, or the engine could not tell) — zero. No verdict, no promise.

The methodology comprises 128 checks.

Only a subset applies to any given site, chosen by its type. The rest are skipped: they take no share of the score and do not affect the final result. A personal site is never offered percentage points for product markup it has no reason to have.

It does not account for your industry, your competition, or your current traffic. It is not calibrated against measured outcomes — doing that would require traffic telemetry we do not collect. Treat the figures as a way to rank fixes against each other, not as a forecast.

Is this a growth guarantee?

No. The impact model is a potential estimate based on published category weights and your current score, not a promise. Every number is labelled as a potential gain and describes the ceiling reachable by fixing the applicable criteria.

Why do different sites get different multipliers?

The base comes from the current channel score: the weaker the site is now, the larger the potential rise. A strong site gets an honest “a potential gain of: +20–100%”, a weak one “a potential gain of: +1500%” (the model’s ceiling).

What does the model not account for?

Competition, seasonality, offline factors and how fast the fixes are applied. It only estimates the effect of removing the AI obstacles found on your site.

Why this many criteria and not three times as many?

Because the list is assembled by influence, not by volume. Adding another hundred checks is easy, but it would not make the score more accurate: it would dilute it across signals that barely change anything for an AI system. We prefer a shorter list where every entry comes with an explanation of what it changes.