Methodology

How we calculate

This page explains how a Brand Score comes about. It exists because the brand check also rates presences that are not ours — and anyone who is rated has a right to redo the maths.

It describes what is running today. If the calculation changes, this page changes with it.

What the Brand Score is

The Brand Score is a number between 0 and 100. It describes how clearly a publicly available web presence expresses its brand. The basis is the public self-presentation: the text every visitor sees, plus the technical declarations in the page head.

It is a classification, not a statement of fact — a reasoned opinion about how a presence comes across from the outside. Every single judgement carries its evidence from the presence itself, so that it can be contradicted.

What it explicitly is not

  • Not a statement about commercial success, revenue or solvency.
  • Not a statement about the quality of the products or services behind it. Excellent craft can present itself badly — and the other way round.
  • Not a statement about integrity or trustworthiness.
  • Not a judgement of taste. We check whether a decision was recognisably made, not whether we like it.

The seven bands

The number turns into a word. The thresholds are fixed and apply to every result alike.

  • Exceptionalfrom 94
  • Outstanding88–93
  • Excellent80–87
  • Strong70–79
  • Average60–69
  • Developing50–59
  • Right at the startbelow 50

What we read

Exactly one page: the address that was entered, after all redirects. No crawling of the site, no second request.

  • The visible body text of that one page, at most 20,000 characters. Anything beyond that is cut off and not scored.
  • The declarations in the page head: title, description, language, favicon, preview image, colour and appearance declarations, structured data, heading hierarchy.
  • The alt texts of images. We do not load the images themselves and we do not store them.
  • Whether the address leads to an encrypted connection.

What we do not read

  • No sub-pages. Whether an imprint, a contact or an about page is linked is something we see on that one page — we do not open it.
  • Nothing behind a login. We do not sign in anywhere and we do not fill in forms.
  • No social media profiles, no review sites, no registers, no press coverage.
  • No JavaScript. We read the document as delivered, not what a browser rebuilds from it.

What is kept

The page text is discarded after the evaluation. What remains stored is the checked address, the numeric values, the short pieces of evidence, the date, and a checksum of the text that tells us whether anything has changed since. Plus a daily rotating checksum of the requesting connection — for the daily cap only, never the address itself.

How to recognise us in your access log

User-agent: PukalaniBrandCheck/1.0 (+https://branding.supply/brand-check/methodik)

robots.txt and mining reservation

Before we fetch a page we read your robots.txt and abide by it — including rules addressed to all bots. We also check for a machine-readable text and data mining reservation (§ 44b UrhG). If either forbids the evaluation, we stop: there is no score, no result and no stored entry — whoever started the check is told exactly that. Until 8 September 2026 this applied only to our market comparison; since then it applies here too.

How to block us

Three ways, each effective with the next check. The first is one line in your robots.txt:

User-agent: PukalaniBrandCheck
Disallow: /

Mining reservation (text and data mining)

The second way: we recognise a machine-readable reservation under § 44b UrhG in these four forms. One of them is enough:

  • As an HTTP header

    TDM-Reservation: 1
  • As a meta declaration in the HTML

    <meta name="tdm-reservation" content="1">
  • As a robots meta declaration

    <meta name="robots" content="noai, noimageai">
  • As tdmrep.json under /.well-known/

    /.well-known/tdmrep.json
    [{ "location": "/", "tdm-reservation": 1 }]

When in doubt we decide against the evaluation: if a reservation cannot be safely ruled out, we stop.

The third way needs no file: write to us and we will block your address on our side. That also applies retroactively to a result that already exists.

The limit of this promise

We ask the origin of the address that was entered. If it redirects to a different host, we do not additionally fetch that host’s robots.txt — our market comparison behaves the same way. And the website that a signed-in operator enters as their own in the brand wizard is still read without this preliminary request: there, the owner of the page is the one giving the order.

How our market comparison bot works

How it is scored

40 criteria in 8 categories. Every criterion has a fixed rule and awards 0, 1 or 2 points. The categories carry different weights; together they make up the 100 points of the overall score.

  • Distinctiveness

    15 of 100 points · 5 criteria

    Does the presence say something only this brand could say — with a stance of its own instead of filler words?

  • Visual identity

    15 of 100 points · 5 criteria

    Is a visual world of its own recognisable, and are the visible marks in place: favicon, preview image, heading hierarchy?

  • Consistency

    15 of 100 points · 5 criteria

    Do title, heading and body speak with one voice, and is the brand name spelled the same way throughout?

  • Brand experience

    15 of 100 points · 5 criteria

    Can a first-time visitor find their way: one clear next action, findable declarations, a recognisable sender?

  • Positioning and clarity

    10 of 100 points · 5 criteria

    Does it say in plain words, without jargon, what is offered, for whom — and what is different here?

  • Emotional impact

    10 of 100 points · 5 criteria

    Does the presence speak to a feeling: an origin, a human voice, a tone that fits the subject?

  • Adaptability

    10 of 100 points · 5 criteria

    Does the brand carry elsewhere too: on a phone, in dark mode, in the sharing preview, in a single sentence?

  • Craft

    10 of 100 points · 5 criteria

    Are the details right: spelling, no placeholder text, an encrypted connection, clean declarations in the page head?

Measured and judged

16 of the 40 criteria are computed deterministically from the document — favicon, preview image, heading hierarchy, language declaration, encrypted connection and the like. The same page always yields the same points there. The remaining 24 are judged by a language model, in one pass, with mandatory evidence for every judgement.

Which model, and under which conditions

We call a large language model through a router — and only providers who have committed to not storing the request. Use for training purposes is prohibited, and falling back to providers outside those conditions is switched off: we would rather have the check say it is currently unavailable. Which model judged a given result is recorded with that result; it can change over time, which is why no fixed name is given here.

Why two runs can differ slightly

The measured part is reproducible. The judged part almost is: we run the model as cold as possible, but a language model is not an adding machine. Above all the basis changes — today's website is not last week's. That is why every result carries its date, and why the same address deliberately returns the same stored result within 7 days instead of putting a second number next to the first.

What cannot be checked from outside, we do not guess

A criterion about which nothing can honestly be said is marked not assessable and drops out of the calculation — it counts neither in the numerator nor in the denominator. Zero points would be a claim about something we never looked at.

Foundation maturity — the second number

Anyone building a branding with us can have their foundation checked instead of a website: the confirmed answers from the wizard. It is calculated with the same formula and the same catalogue of criteria.

That is why this number has a different name and is labelled separately everywhere. The 16 measured criteria are all not assessable there — a favicon and a heading hierarchy are properties of a page, not of a foundation. A foundation maturity of 78 and a website score of 78 say different things; the side-by-side view refuses to compare across that line.

Foundation maturity exists only for your own brand. It cannot be established for someone else's presence.

What is public — and what is not

A result is private to begin with

A check enters our public ranking only if the person running it explicitly ticked the box at the start. Without that tick it appears in no list, no sorting and no ranking position.

The result page itself

Every result gets its own shareable address. Anyone who knows it can open the result — it is meant to be passed on. Search engines do not index it, and it shows nothing that would not be public anyway: the checked address, and judgements about a publicly reachable page.

We are aware that the person running a check is not always the brand owner

That is why there are two ways back. Both are free and need no account.

Suggest a correction

A correction can be submitted for any result: the field concerned, the suggestion, and a reason. A contact address is optional. We review by hand and either accept or decline with a reason. What can be corrected today is the industry — it is a model's suggestion about someone else's presence. The score itself is not contradicted by a correction, but through the route below.

3 suggestions per hour and connection.

Have it removed

On a substantiated request we take a result offline. It disappears from the ranking and from its own address as well — what stands there afterwards is the same as for a result that never existed. The removal is permanent: it is not undone by someone else checking the same address again.

How to reach us

For a correction the form on the result is enough. For a removal, to have your address blocked, or to query a judgement, write to us using the contact details in the imprint.

Or by e-mail to hello@branding.supply

To the imprint

Limits and fairness

  • A model can be wrong

    The judged criteria are a machine judgement. It carries its evidence and is therefore verifiable, but it is not infallible. If you see a mistake, tell us.

  • The score is a snapshot

    It holds for one page on one day. A relaunch, a new text, a changed home page — and the number no longer holds. That is why every result carries its date, and why an old value is not a valid value to us.

  • We do not pit two brands against each other

    There is a side-by-side view, but it is descriptive: two results next to each other, category by category, with no winner and no disparagement. We do not claim that one brand is better than another.

  • No third-party logos

    We show word marks, industry and short supporting quotes from the presence itself. No logos, no figurative marks, no favicons, no screenshots.

  • A home page is not the whole brand

    A brand also lives in the product, in the shop, in conversation and in service. This check sees none of that. It measures the first impression online — that matters, but it is not everything.

To the brand checkTo the ranking