SuppScore · Phase 1 Foundational Sprint · Prepared by Whirlwind

The Architecture Blueprint

The pages already designed are good. What is missing is the structure that ties them together and lets the site grow without anyone designing a new page. This document sets that structure, and ends with four decisions only you can make.

Deliverable 1 of 5 Pre-launch UK English at launch, built to travel
4layers, each one feeding credibility to the next
11page templates at launch, one more held in reserve
6of them built automatically from records, with no design work
4decisions we need from you before the next deliverable

Start here

Why the blueprint comes first

The homepage, the ingredient dictionary, the product listing, How We Score: the right instincts are all there. The audit found five things missing between them, and every one of them is a structural decision rather than a bug.

The whole site shares one page title

Every page, from the homepage to the magnesium entry, carries the same title and the same description. To Google, the site currently looks like one page.

Products have no name in their web address

Every product sits at /results?scanId= followed by a random string. Nothing in that address tells anyone, or any search engine, which product it is.

Benefit categories are not pages

Sleep, Energy and Immunity are filters on the product list. The main way we want people to navigate the site has no web addresses of its own.

There is nowhere for a brand to land

People search "is Dr Vegan any good" every month. Today that search has nowhere to go on SuppScore.

Nothing is readable by machines

No structured data on any page, no robots file, no sitemap. Nothing yet tells Google or an AI assistant what a score is, what it is made of, or which pages exist.

None of these are bugs. They are what happens when pages get built before the structure that connects them is agreed. Every one is close to free to fix now and expensive to fix after launch, when fixing means moving addresses, setting up redirects, and re-earning credit that was already built up against the wrong pages.

The stakes are specific to this market. SuppScore launches with no authority into a category owned by Examine and the NIH, and 80 to 90 percent of the biggest supplement searches in the UK now show an AI answer at the top. We cannot outspend the incumbents on authority. Structure is the one thing we fully control, and it gets set in the next few weeks or not at all.

What this document delivers. A four-layer structure, eleven page templates for launch with one more held in reserve, and a set of publishing rules that let the site grow from 500 products to 50,000 without anyone designing a new page.

The structure

Four layers

Trust makes the knowledge believable. Knowledge makes the demand pages worth reading. Demand pages send people to products. Products feed data back up.

How the site is organised
Each layer borrows credibility from the one above it
CREDIBILITY FLOWS DOWN Trust Where the site's credibility is created. Every score sits one click from here. Methodology Advisory board Company and policies Knowledge The reference library AI assistants quote. The part nobody else has. 145 ingredient pages 80 to 150 forms 137 additives Demand Where searches land. Built from the scoring data, not written from scratch. 25 benefit pages Top-20 lists 20 to 40 question pages Comparisons Product The biggest source of traffic as the database grows, and it grows on its own. Product pages Brand pages

The page types

The template library

Every page on the site is one of eleven templates. One more is designed and held back for later. Six are built automatically from records in the database. The rest are either built only where we can see people searching, or built by hand.

1Ingredient page

Built automatically

The main page for each ingredient, and the page everything else about that ingredient points to.

Wins: "magnesium", "magnesium benefits", "magnesium dosage"
/ingredients/magnesium

Why it earns its place. Ingredient pages are 72 percent of Examine's top 1,000 UK searches. The NIH holds 467 top-three positions with the same page type. The single most valuable page type in the category.

2Ingredient form page

Built automatically Where searches exist

For the specific forms people search by name.

Wins: "magnesium glycinate", "magnesium glycinate benefits"
/ingredients/magnesium/magnesium-glycinate

Why it earns its place. Examine's own magnesium glycinate page returns an error. These searches are answered today by shops and blogs. Our 683 scored forms are the only dataset that can answer them properly.

3Additive and excipient page

Built automatically

The other half of the score, explained. The fillers, binders and coatings we mark products down for.

Wins: "is magnesium stearate safe", "titanium dioxide in supplements"
/ingredients/magnesium-stearate

Why it earns its place. We already hold around 137 scored excipients. No site we studied grades additives at all.

4Benefit page

Hand-built, around 25

The main way people navigate the site, and the top of each section.

Wins: "sleep supplements", "supplements for energy"
/benefits/sleep

Why it earns its place. Examine runs 26 health categories. Our planned 20 to 30 is the right size, and it is the navigation path already confirmed.

5Top-20 list

Where searches exist

The commercial page. States a winner, a score and a reason, and says so in its own page description.

Wins: "best magnesium supplements", "best sleep supplements UK"
/benefits/sleep/best

Why it earns its place. Ranked list pages account for 97 percent of Labdoor's search visibility. This is the page type the pre-launch ad plan already depends on.

6Question page

Hand-picked, small set

For questions that need a real answer, not a paragraph. Governed by the rule on the next screen.

Wins: "does magnesium help with sleep"
/qa/does-magnesium-help-with-sleep

Why it earns its place. Question-shaped content is quoted by AI assistants far more often than its share of rankings. Examine's question pages are 7.5 percent of their rankings but 17.5 percent of their AI citations.

7Product page

Built automatically

A label decoded, with a score.

Wins: "{product} review", "{product} ingredients"
/products/dr-vegan-magnesium-glycinate

Why it earns its place. This page type earns IncideCoder 2.27M monthly searches and accounts for 474 of Unbox Health's 515 ranked searches.

8Brand page

Built automatically Three products or more

An honest summary of everything we have scored from one brand. Not a navigation path.

Wins: "dr vegan reviews", "dr vegan products", "is dr vegan any good". Not "dr vegan", which their own site will always win.
/brands/dr-vegan

Why it earns its place. Real demand with nowhere to land today, and exactly the kind of search an independent scorer can win. Three rules: the page links out to the brand's official website and marks it as the official source, so nobody thinks we are pretending to be them; the page reports the data and passes no judgement on the company; and it is only created once we have scored three or more of their products, so it has something to say.

9Form comparison

Where searches exist

Direct comparisons carrying our own verdict: the score difference, the dose difference, and a plain answer.

Wins: "magnesium citrate vs glycinate"
/compare/magnesium-citrate-vs-magnesium-glycinate

Why it earns its place. Worth having, easy to ruin. 83 percent of Unbox Health's sitemap is comparison pages earning 3 percent of their traffic. These get built from the keyword research only, never by pairing every form with every other form.

10Guide

Hand-built

The campaign page. Pre-launch Top-20 articles, ad landing pages, PR pieces.

Wins: editorial and paid traffic, not a standing search position
/guides/magnesium-forms-explained

Why it earns its place. This is what the pre-launch plan of 500 products plus Top-20 articles plus paid traffic already needs.

11Trust pages

Hand-built, small set

Methodology, about, advisory board, policies.

Wins: the credibility every other page borrows
/how-we-score

Why it earns its place. Labdoor describes its method but never publishes its weights or its formula. Publishing ours in full, in a form machines can read, is the clearest difference between SuppScore and everyone else.

12Ingredient in benefit

Held back, not at launch

Pages like "magnesium for sleep". Real demand exists, but at launch, with 145 ingredients and 500 products, we could not write anything on those pages that is not already on the magnesium page and the sleep page.

/benefits/sleep/magnesium

What happens instead. Those questions get answered as FAQ sections on the ingredient page, linked across to the relevant benefit page. The web address slot stays reserved, so when the database is deep enough to justify these, it is a new page type dropping into a space that already exists, not a rebuild.

Plus one tool that is deliberately not a template: the analyser itself. Results from a pasted ingredient list stay on /results and stay out of Google.

The rule that protects the site

Keeping question pages from taking over

This is the page type that most often wrecks sites like this, so the rule is worth stating on its own. Left alone, a database of 145 ingredients and 25 benefits can generate thousands of question pages. Most would be one paragraph long, and Google treats a site full of one-paragraph pages as a low quality site. That judgement then affects the good pages too.

The ruleA question gets its own page only when the honest answer is too long to sit on the parent page.
Everything elseBecomes an FAQ section on the ingredient page or the benefit page it belongs to. The same question never appears in two places, because two pages answering the same question compete with each other and both lose.
The test before a page is createdCan we write 300 words that a specialist would recognise as correct, with at least two references and a named reviewer? If not, it is a line in an FAQ.

At launch that means roughly 20 to 40 question pages, chosen from the keyword research in Deliverable 2, not generated from a list. It also means the FAQ sections on ingredient and benefit pages matter more than they look. They are the same question-and-answer format that gets quoted by AI assistants, sitting on pages that are already strong.

The decision the knowledge library rests on

The forms decision

Option A: everything sits on the parent page

One magnesium page. Glycinate, citrate, oxide and the rest each get a row in a table and a line in an FAQ. This is what Examine and the NIH do.

Safe, and it can never produce a thin page. It also leaves every search for a specific form, "magnesium glycinate benefits", "glycinate dosage for sleep", to be answered by shops and blogs. We would own the best form-level data in the category and never show it.

Option B, our recommendation: parent page first, best forms promoted

The magnesium page stays the main page. It carries a summary row for every form: tier, multiplier, one-line reason, reference.

A form gets its own page only when two things are both true. People are searching for that form by name, and our data holds something distinct to say about it. Promoting a form is a setting on a record, not a piece of development work.

The testReal searches plus something distinct to say, or the form stays a row on the parent page.
Why it is worth doingThe form level is the one part of this category nobody currently occupies, and our 683 scored forms are the only dataset that can occupy it without producing thin pages.

We expect 80 to 150 forms to get their own page at launch. The exact list comes out of the keyword research in Deliverable 2, not out of anyone's opinion.

Why this survives scale

How it grows

When your team adds ingredient 146, or the database reaches product 5,000, nobody designs anything.

One new record, no new design work
The same path runs for every ingredient, product, additive and brand added from now on
New record in Ingredient 146. Product 5,000. Template applied Own title, one heading, breadcrumbs, score as text Links itself in To its benefits and its products, both ways Published Complete, real demand Held back Kept out of Google Error page Address is not real

Three publishing rules decide what Google sees. A complete record with real search demand gets published. A record missing required information still renders for users but stays out of Google until it is complete. A web address that does not match a real record returns a proper error page, never an empty page pretending to be real.

The same applies to categories. Benefit categories are records, not designs. Adding sports nutrition or kids' supplements later is a new record inheriting the benefit template, not a redesign. Deliverable 2 maps search demand against whatever categories exist at the time.

Why the rules are not optional. Open Food Facts has millions of pages in Google saying "this page is not complete" and survives only on reputation we will not have. Unbox Health returns a real-looking page for any made-up address. These rules are what let SuppScore accept crowdsourced growth without inheriting crowdsourced junk.

The end state

What it looks like when it works

Three moments, all direct results of the structure above. Worth two sentences of context first: the established sites are shutting AI systems out, with drugs.com alone blocking around 20 AI crawlers and ConsumerLab sitting behind a paywall. Only one site in our whole study explicitly lets AI in.

Someone asks their assistant about magnesium

Our robots file welcomes the assistant's crawler by name, our firewall never blocks it, and the page's raw code carries the tier, the multiplier, the dose zones and the references as plain readable text.

"For sleep, magnesium glycinate is generally the better-absorbed form."Source: SuppScore

Winnable immediately, because being quoted by an assistant depends on being readable, not on being famous.

Someone googles "best magnesium supplements UK"

The Top-20 page states the winner, its score and the reason in its own page description. When Google's AI answer quotes a source, the sentence it lifts is ours, with the grade attached.

Top magnesium supplements in the UK, scored. Highest rated: 8.4 out of 10, Tier A form at a clinically studied dose./benefits/sleep/best

Slower, because it needs rankings first. Same page, both jobs.

Someone is shopping on a retailer's website

A SuppScore badge sits on the product page, linking back to that product on our site. Every badge is a link, a mention and a reminder at the exact moment someone is deciding what to buy.

SuppScore 8.4 / 10links to /products/dr-vegan-magnesium-glycinate

A later decision, but the page it links to is being built now, and the address will not change under it.

The point of this document

Four decisions only you can make

Everything above is settled. These four are not, because they are judgement calls about the business rather than decisions about structure. Two pairs of them are connected, and it is worth seeing how.

1How far do we go on dosage? Decided by 4

The question
Do we describe what doses the research has used, or do we stay strictly on quality and scoring?
Why it matters
Dosage is the biggest question people ask in this category, and it is hard to explain a low score for a 50mg magnesium product without saying what dose the research actually used. Staying silent means publishing a number we will not explain.
Where the line sits either way
No drug interactions, no treating conditions, no "signs you are deficient", and never personal advice. One standing safety note sitewide telling anyone on medication or with kidney problems to speak to their doctor.
Our position
Describing published dose ranges, with references and a named reviewer, is the strongest thing we own. It only works if decision 4 is a yes. Without named, credentialed reviewers on those pages we would be taking the risk without the credibility, and the safer answer becomes the right one.
Cost of deciding late
The question pages and the keyword research both split on this answer. Deciding after Deliverable 2 means redoing part of it.

2Do we publish a corrections and disputes policy at launch? Pairs with 3

The question
When a brand disagrees with a score, what happens?
Why it matters
This is the professional answer to the first angry email, and it is a trust signal in itself. Nobody in this category has one.
What it actually is
Two paragraphs and an email address. If we got a fact wrong, the label says 150mg and we published 100mg, we fix it fast and the score updates. If someone disagrees with the method itself, that conversation happens in public, because the method is public.
Our recommendation
Yes, at launch. One page, linked from every score.
Cost of deciding late
The first dispute gets handled off the cuff by whoever opens the email, in public.

3When scores change, do people see that they changed? Pairs with 2

The question
As the method improves and products get re-scored, do scores update quietly or visibly?
Why it matters
A score that changes quietly looks like something was arranged, the first time anyone compares two screenshots. A dated update history is how a serious reference behaves, and it is what makes the corrections policy above believable.
Our recommendation
Dated, visible updates shown on the page.
Cost of deciding late
Adding a history later means the earliest changes have no record, which is exactly the period anyone challenging us will ask about.

4Will the advisory board go on the record? Decides 1

The question
Will board members put real names, real credentials and links to their professional profiles on the site?
Why it matters
This is health information. Google and AI systems both trace trust back to named people who can be verified elsewhere on the web. The About page currently has placeholder names.
Our recommendation
Yes, and we would treat this as the one non-negotiable. Named reviewers on the methodology and on ingredient pages.
Cost of deciding late
Every page published without a named reviewer is a page that has to be updated and re-crawled later. It also decides decision 1.