Quality & Evidence — how we make sure everything is right

Plain-English briefing on the verification, checking and evidence layers that run through every stage of the platform — the answer to "how do I know I can trust this?" Written 24 July 2026. Last verified: 3 September 2026 (Agent 11 — Co-Kinetic Sites "Reviewed by" display folded in).

Why this document exists

AI-generated content has a deserved reputation problem: fabricated references, confident errors, generic filler, and no one accountable for any of it. A clinic putting its professional name on published health content cannot afford any of those failures — and neither can we.

So the platform is built the other way up. Rather than generating content and hoping it's right, every stage — from the first assessment measurement to the final published page — has its own verification layer. This document walks through them all in one place. The short version: nothing is asserted that hasn't been checked, nothing unverifiable is silently passed, and nothing is ever fabricated.

Where our quality standard comes from

This isn't a software company's first encounter with clinical accuracy. For 25 years and 100 issues, Co-Kinetic published the Co-Kinetic Journal (formerly sportEX Medicine) — a peer-reviewed clinical journal for musculoskeletal and sports medicine professionals. A quarter of a century of commissioning, editing, peer-reviewing and publishing evidence-based clinical content for a professional readership teaches you exactly what clinical rigour costs, what corner-cutting looks like, and why "roughly right" is never acceptable when practitioners put their names to it.

That editorial DNA is built into the platform. The verification layers described in this document aren't compliance box-ticking bolted onto an AI product — they're a peer-review publisher's standards, encoded into software. Very few companies building AI content tools have ever run a clinical journal; it shows in what we check.

The foundation: measured evidence, not opinion

Everything the platform recommends traces back to something measured:

• The assessment crawls the clinic's entire website and evaluates every single page individually — not a homepage skim, not a sample.
• It draws on multiple independent external data sources: live probing of AI search engines themselves (does AI actually cite this clinic today, for realistic patient questions?), Google Business Profile review data, an independent search-results provider, an independent search-demand provider, and a deep-research capability for anything automated collection can't reach. No single tool's opinion decides anything.
• The quality standard every page is measured against is grounded in published research into how AI search engines select and cite sources — not our hunches, and not recycled SEO folklore.
• Where sources disagree or something can't be established, the platform says so — it flags the item, triggers a notification, and brings a human into the loop to resolve it. "Needs verification" is an honest state; a silent guess is not allowed. Automation does the heavy lifting, but ambiguity always escalates to a person.

The research behind every article

Every piece of content starts from fresh, structured clinical research — never from an AI model's memory:

• Research runs through a live research engine with real-time access to current sources, gathering clinical evidence, statistics and references for the specific question the page will answer.
• The research process is structured — up to 12 defined steps that gather, cross-reference and consolidate the evidence before a word of the article is written. The more significant the content, the deeper the run goes. Core steps include patient voices (how real patients describe the condition — the questions, language, fears and misconceptions AI search is actually being asked about), the clinical evidence base (the current state of what works, honestly including where evidence is limited), and clinical citations (the specific published research supporting the article's claims, gathered with verifiable identifiers).
• Research is banked and kept current: once a topic has been properly researched, later content builds on it — with anything stale refreshed rather than blindly reused, and the library refreshed on a rolling basis so the evidence underneath the content stays up to date. Quality compounds; staleness doesn't.
• Research is framed by the author's actual profession — a physiotherapist's, osteopath's, chiropractor's or massage therapist's page each reflects their discipline's genuine approach, because a page describing the wrong profession's methods under a practitioner's name would be a trust failure, not a shortcut.

Citations: real, verified, and pointing at the right paper

References are the highest-stakes element of clinical content, so they get the deepest checking in the platform:

1. Every cited reference must be real. Journal citations carry DOIs (the publishing industry's permanent identifiers), and every DOI is checked against the global DOI registry at the moment of generation. A reference that doesn't resolve to a genuine registered publication is structurally removed — it cannot reach the page. This closes off the single most notorious AI failure: the plausible-looking, entirely invented reference.
2. Verification happens at multiple points, not just once: at generation, again when banked research is reused (so a bad reference can't sneak back in through the library), and again at the moment the page is served.
3. Beyond "real" to "right": our content-health monitoring additionally checks that each reference resolves to the paper the article says it is — catching the subtle failure where a genuine registered DOI points at a different publication than claimed. To our knowledge, almost nobody else checks for this at all.
4. Anything that can't be positively verified is flagged for human review — never silently passed as fine.

Quality gating: checked before the customer ever sees it

Every generated article is scored against the platform's full quality standard before delivery — the same standard used to score the clinic's existing website in the assessment, so one consistent bar runs through everything:

• An article scoring below the quality threshold is automatically repaired and re-scored, and if necessary regenerated — a self-healing loop that runs before the customer sees anything. The customer reviews finished work, not drafts.
• Articles that needed healing are tracked internally, because a pattern of struggling output points at an upstream improvement we should make — the quality system feeds its own refinement.
• Every article's quality score is visible to the customer on the piece itself. Quality is never "trust us."

Continuous monitoring after generation

Checking doesn't stop at delivery. An automated content-health layer verifies every generated piece on an ongoing basis — structural integrity, citation hygiene (including the destination-correctness check above), and the correctness of the machine-readable data attached to each page. Failures surface to our team with severity grading; anything ambiguous routes to a human rather than being auto-passed.

Human oversight where it matters most

Automation does the heavy checking; people hold the clinical accountability:

• Every article is authored under a named, registered practitioner who reviews it before publication and can edit anything.
• An optional peer-review step lets a second clinician sign off, with their credentials attached to the published piece.
• The practitioner's own clinical insight — collected from them directly — shapes the article, so the clinical perspective is genuinely theirs, not synthesised.

Honest numbers, always

Two platform-wide rules protect every number a customer or their patients ever see:

• No fabrication, anywhere. A clinic with no reviews shows no rating — never an invented one. A page that can't be verified isn't marked fine. A statistic without a source doesn't appear. The same rule extends to professions themselves: when the platform researches a profession it hasn't seen before, regulatory bodies are named only where they can be verified to genuinely regulate that profession in that country — where they can't, the content says so honestly rather than guessing.
• Benchmark history is never destroyed. Every assessment writes a dated snapshot that is preserved permanently, and improvement is measured against fixed benchmarks — so when a clinic's score moves from 52 to 67, that is a genuine like-for-like gain, evidenced in their own preserved history.

The whole picture

Every row in that table exists because we assumed something would go wrong there and built the check before it could. That's the design philosophy in one sentence: assume nothing, verify everything, fabricate never.

Common questions this briefing answers

• "Where does the content get its information?" From a structured clinical research process of up to 12 steps, run against live current sources at the time of writing — never from an AI's memory — covering patient voices, the clinical evidence base and verifiable clinical citations, with the results banked, freshness-managed, and refreshed on a rolling basis.
• "How do I know the references are real?" Every one is checked against the global DOI registry before it can appear, re-checked on reuse and at serve time, and monitored afterwards for pointing at the right paper. Unverifiable references go to human review, never silently through.
• "What stops it publishing something wrong about my profession or my clinic?" Content is anchored to the author's actual professional profile and discipline, the clinic's real services and country context — and the named practitioner reviews and can edit everything before it's published.
• "Why doesn't my website page show who reviewed it?" (Co-Kinetic Sites customers.) The "Reviewed by [clinician]" line appears automatically on every condition, treatment, service and blog page once the article's peer review is APPROVED in CCS (live 28–30 Aug 2026). A page with no reviewer line means the review isn't approved yet — nothing to configure; approve the review and the line appears.
• "Who checks the checker?" The quality gate's own activity is tracked, the content-health layer monitors published output continuously, and our team reviews everything those systems flag. And the customer's own score history — measured on fixed benchmarks by independent instruments — is the final audit no one can massage.