Research & Content Generation: how the content gets made, and why yours is genuinely different

Updated 22 October 2026 (Agent 11.1: patient examples rule, reviewer checks examples; change feed folded in: regenerate replaces the draft, #3877; single author card). Previously updated 11 October 2026 (Agent 11.1: reviewer step and Wix publishing steps, #3711); earlier: Updated 7 October 2026 (Agent 11.1: catch-up from every lane after their state-doc updates; scoring rebuild of 21 Sep, no score-trend claims, no citation claims for schema or credentials, Sites, content plan, support). Plain-English briefing on the five-stage pipeline from plan action to finished article: with particular attention to personalisation, because "won't my content be the same as everyone else's?" is the most common misconception about AI-generated content. Written 24 July 2026. Updated 18 August 2026: the fresh practice reading before every article, and how we measure sameness rather than assert its absence. Last verified: 24 September 2026 (Agent 11, fixed criterion count removed per the rubric rebuild). Previously: 27 August 2026 (Agent 11, AHPRA "specialist"-wording tightening folded in).

The short answer to the big objection

"If it's AI-generated, won't my content be the same as every other clinic's?"

No, and not because we say so, but because of what goes into it. The article written for your clinic is built from: research targeted at your page's specific question and framed by your practitioner's professional discipline, your named practitioner's real credentials and registration, your practitioner's own clinical insight (their words, collected from them), your clinic's services and locations, your local community context, your country's clinical and funding framework, and your existing site's pages (which the article links to). None of those inputs exist for any other clinic. Two clinics asking for "a neck pain page" diverge at every stage of the pipeline, a different author with different professional expertise, different insight, different country framing, different local context, different internal links, a different set of services shaping what the page recommends next. Generic AI content is what you get when you paste a prompt into ChatGPT. This is the opposite: a pipeline where the differentiating inputs are structural.

The five stages

Stage 1: Research

Every article starts with fresh, targeted clinical research, not the model's memory.

• Research is per-page-type: a condition page, a treatment page, a service page and a blog post each have their own research pipeline asking different questions, because a patient reading "what is chronic Achilles tendonitis?" needs different evidence than one reading "where can I get treatment?"
• Research is a structured process of up to 12 steps, scaled to the significance of the content: the more important the page, the deeper the run. Among the core steps:
  • Patient voices, how real patients actually describe this condition: the questions they ask, the words they use, their fears, misconceptions and what they wish they'd known. This step matters enormously, because AI search is answering patients' natural-language questions, content built from how patients genuinely talk is precisely what AI engines select to answer them.
  • Clinical evidence base: the current state of the clinical evidence for the topic: what works, what doesn't, where the evidence is strong and where it's honestly limited.
  • Clinical citations: the specific published research that supports the claims the article will make, gathered with their identifiers so every reference can be verified (see the citations section below).
• Research runs through a live research engine with real-time web access, gathering current evidence, statistics, and sources at the time of the run: never from an AI model's memory.
• Research is banked in a research library and kept current: subsequent content on a topic checks the library first, reuses what's fresh, and re-runs what's stale, and library research is refreshed on a rolling basis so the evidence base underneath the content stays up to date, not frozen at the moment it was first gathered.
• Research is anchored to the author's profession, and this covers the full breadth of practice: physiotherapist, osteopath, chiropractor, sports massage practitioner, or allied discipline. An osteopath's page on a condition reflects osteopathic assessment and treatment framing; a physiotherapist's reflects physiotherapy, because the professions genuinely differ in approach, and a page that described the wrong discipline's methods under a practitioner's name would undermine their credibility rather than build it. The profession flows from the author's profile through research framing, terminology, treatment descriptions, and regulatory references: right through to the writing stage, where the author's professional identity is an explicit instruction, not something the system infers.
• And the set of professions is open, not fixed. When a practitioner with a profession the platform hasn't seen before joins a clinic, a Bowen practitioner, a sports kinesiologist, a veterinary physiotherapist, the platform automatically researches that profession in the clinic's own country: its scope of practice, the conditions it treats, its natural patient-facing language, and its regulatory position there, with verified regulator names only, never guessed; where a profession has no statutory regulator in a country, the research says so honestly rather than inventing one. This sits alongside what the practitioner declares themselves: every clinician records their own professional registrations and memberships on their profile, and those are what appear against their name in bylines and structured data, so the regulatory picture around their content is grounded in both the practitioner's own declared registration and the platform's independent verification of the profession. The profession is then represented correctly in content from that point on, with no manual setup and no delay to the clinic. Until the research completes, wording degrades gracefully to neutral clinical language rather than defaulting to any particular profession.
• Research also injects the practice profile and local context directly where the page type calls for it (blog research in particular).

Stage 2: Author and insight

Every article has a named author: a real practitioner at the clinic.

• The platform suggests the best-matched author from the team based on declared niches and CPD relevance; the clinic confirms or overrides.
• The author is invited to contribute clinical insight through a short structured form: what's the biggest misconception patients have about this? What's your treatment philosophy? What do you wish patients knew? Their answers are woven into the article as genuine first-person clinical perspective: the single most inimitable ingredient in the pipeline.
• Patient examples: what we ask for, what we never ask for (agreed 22 Oct 2026, same in every country). We ask the clinician for what they usually see with a condition, or an anonymised example from their own practice: the kind of person, what they found and what the care involved. We never ask for a name or anything that identifies the patient, a patient quote, or how it turned out for that patient. Why: we never invent examples, so it must come from the clinician; and in Australia and New Zealand a patient's result counts as a testimonial, which health advertising doesn't allow. We apply the same rule everywhere so every article is safe in any country. The writer only uses an example the clinician supplied, and never states a result even if the clinician's text gives one. Plans never suggest patient stories, case studies, testimonials or success stories as pieces. Customer help: "Can I include a patient example in my article?"
• Insight is never lost: once a practitioner's insight is on file for a piece, every future regeneration of that piece automatically re-weaves it. The clinician contributes once; the value persists.
• A second clinician acts as reviewer, and we recommend one for every article. The clinic picks the reviewer, the reviewer gets a review request by email straight away and approves from the link, and their name and credentials then show as "Reviewed by" on the page and in its structured data. Until then the article shows that it is waiting for the reviewer to approve; older articles have a Send button to send the request. A named, qualified reviewer is an important trust signal for patients, search engines and AI tools (trust wording only; never claim a reviewer gets a page cited more). The reviewer is also a second pair of eyes on any patient example: if anyone could recognise the patient, or it says how it turned out, it comes out.

Stage 2b: A fresh reading of your practice, before a word is written

Since August 2026, every article generation begins with something no template system can do: the platform reads your practice fresh, every single time. Your clinicians and what they actually treat, their professional development and recorded interests, the clinical insights your team has given on earlier articles, your services, and even what your clinicians do beyond the clinic, coaching a junior team, pacing a parkrun, teaching on a degree course, all of it is read at the moment your article is created, never from a stored copy and never shared with any other clinic.

That reading then shapes the article before writing starts, which evidence deserves emphasis for the patients your clinic actually sees, and which patient questions your article should answer first. A clinic whose team lives and breathes running injuries gets a runner's-knee page shaped around the questions runners ask; a clinic focused on older adults gets the same clinical facts shaped around a different life. And the reading that shaped each article is stored with it, so if you ever ask "why does my article emphasise this?", the answer is on record.

One rule always wins, for your protection: clinical safety and advertising compliance outrank everything. However distinctive your practice's material, nothing is ever rendered as a patient story we invented, a patient quote or an outcome claim. An anonymised example appears only if the clinician wrote it, and never with a result.

Stage 3: Writing

The writing stage assembles everything into a finished article, structured for how AI search actually reads:

• Answer-first structure (the key answer in the first paragraph, where AI engines look for quotable summaries), clear headings, an at-a-glance facts table, an FAQ section.
• A byline card at the top: author photo, name, credentials, professional registration number, link to their full profile. Reviewer credit at the end where applicable.
• Real citations only. Named journals with DOI links. Every DOI is verified against the global DOI registry at generation time: a citation that doesn't resolve to a real paper is structurally removed, not published. Fabricated references, the classic failure of naive AI content, are prevented at the infrastructure level (verification at generation, at research-reuse, and again at render).
• Country-aware throughout. An Australian clinic's article references Medicare Chronic Condition Management, NDIS and WorkCover, and AHPRA-registered practitioners; a UK clinic's references NHS pathways and NICE guidance. The country flavours the substance naturally rather than being stamped on as a label. (Since 20 Aug 2026, AU generation is tightened for AHPRA compliance: the protected title "specialist" is avoided in every form, "specialises in", "specialising", "specialty", across all AU article prompts and the AU team-member page prompt; clinicians without specialist registration are described as having a "special interest" or "practice focus". Existing published pieces keep their old wording until regenerated, and customers remain responsible for the final read of AI-generated content.)
• Multidisciplinary clinics get multidisciplinary content. Where a practice's active team spans more than one discipline, say osteopathy and sports massage, condition and treatment articles can include a short "Different approaches at [clinic name]" section explaining how each of the clinic's own disciplines approaches the topic, what a patient would experience with each, and how they complement one another. Each discipline is described accurately in its own terms, never ranked against the others, and, the hard rule, no discipline the clinic doesn't actually offer is ever mentioned. Single-discipline clinics see no change: their content stays purely in their own discipline's framing.
• Rooted in the local community. The assessment discovers the clinic's local entities, the running clubs, cycling groups, sports teams, gyms and community organisations around the practice, and where there's a natural opportunity, the content weaves them in: a knee-pain page from a clinic near a well-known parkrun reads like it belongs to that place, because it does. This is a personalisation layer generic content cannot fake: you have to actually know the area.
• Machine-readable structured data generated alongside the visible article: author/organisation/medical-page schema, breadcrumbs, FAQ schema, and, where the clinic has real reviews, a genuine aggregate rating. This is the layer AI engines parse directly, and it's built and validated automatically.
• Length discipline: over-long drafts are automatically condensed to the page-type's target while preserving the byline, tables, citations, insight and schema, then re-scored to confirm condensing cost nothing.

Stage 4: The quality gate

Before the customer ever sees the article, an automated pre-delivery quality gate scores it against the same research-grounded checklist used in the assessment (no fixed criterion count; see 09). Below-threshold output triggers automatic repair passes and, if needed, full regeneration, a self-healing loop that runs without human intervention. The customer sees finished work, not drafts. (Internally, even the gate's own activity is tracked, so a piece that needed healing is flagged for pattern analysis, a systematically struggling page type points at an upstream improvement.)

Stage 5: Review and publish

The finished article lands in Content Studio, the customer's content workspace, moving through visible states: Written → Review → Ready to publish → Published. The customer can read, edit (with full editor), send to their reviewer for approval, and then publish via the routes described in the publish briefing. Every article also carries its own AI-visibility score, visible to the customer, so quality is never a matter of trust.

Ongoing verification: content health

Generation isn't the end of scrutiny. An automated content-health layer verifies every generated piece: prompt versions used, structure integrity, citation hygiene, including that every cited reference not only exists but resolves to the paper the article says it is (a registered DOI pointing at the wrong paper is caught, which is a failure mode nobody else even checks for). Anything unverifiable is flagged for human review rather than silently passed.

"Will my content be different enough?": measured, not promised

This is the question behind most others, so here is the honest, complete answer.

Covering the same condition as another clinic is fine. Every physiotherapy practice in the country writes about back pain, and search engines expect that. The clinical facts should agree: the evidence is the evidence, and a page that bent the science to sound different would be worse, not better. What actually hurts a website is near-identical text: when two sites publish almost the same article, search engines pick one to show and quietly ignore the other.

So we don't write from templates, and we don't just claim that: we measure it. Every published article on the platform is compared, paragraph by paragraph, against every other clinic's published content on the same kind of topic. We built this measuring instrument before we even had two clinics whose content could collide. And we stress-test ourselves deliberately: we run the same topics through the pipeline for deliberately different test clinics and measure how alike the results are, every time we change how the system writes.

The result of all this, as of August 2026: with the fresh practice reading in place, we have measured that different clinics' articles on the same topic are meaningfully more distinct than before, and the improvement is concentrated exactly where similar content tends to converge, the frequently-asked-questions sections, with no drop in clinical quality. The parts that rightly stay similar are the clinical facts. The parts that make a page yours, who it speaks to, whose expertise stands behind it, which questions it answers first, the local texture, come from your practice and can't be copied by another clinic signing up to the same software, because they don't have your people.

And if two articles ever did drift too close, the measurement would show us the exact matching paragraphs, and the piece would be rewritten from that clinic's own facts. Sameness on this platform is a monitored number, not a hope.

Common questions this briefing answers

• "Won't it sound like AI?" The structure is optimised for AI search, but the substance is clinic-specific: real research, a real author's insight, real credentials, real local context. It reads like a well-organised expert page, because that's what it is.
• "Are the references real?" Yes, verifiably: every DOI is checked against the global registry at generation time, and the content-health layer additionally confirms each reference resolves to the claimed paper.
• "Could the references in my article ever change after I've received it?" Occasionally, yes, for the better. Automatic reference checking verifies every citation (real DOI, the right paper, supporting the claim it's attached to) and fixes or removes any that fail. If an article you've already pasted onto your own site is improved this way, re-pasting it brings the corrected references onto your live page. Articles pasted before mid-August 2026 may carry older reference links until re-pasted, if you're unsure whether yours is affected, just ask here.
• "I regenerated an article; why is there only one draft?" Regenerating replaces the draft in place, so no duplicate appears (#3877, live 17 Oct).
• "Why does my article have two author boxes?" An older step could add a second author card of its own. That stopped on 22 Oct, and the pages already affected are being repaired. There is only ever one byline card, from the clinician's profile.
• "What if I want to change something?" Full editing in Content Studio, and edits are respected: the platform warns before any regeneration would overwrite customer edits.
• "What does the practitioner actually have to do?" Minimum: nothing (the pipeline runs without insight). Recommended: five minutes answering three insight questions: the highest-leverage five minutes in the whole process.
• "What happens if the clinic down the road signs up as well?" Their content still won't read like yours, because your content isn't made from a template: it's made from your clinic: your clinicians, their expertise and interests, your reviewer's name on every page, your local texture. A neighbour on the same platform gets articles anchored to their people, not yours. And unlike anywhere else, the overlap between the two sites is something we actively measure, not something anyone has to take on trust.
• "Will Google penalise us for duplicate content?" Covering the same conditions as other clinics is normal and expected: the risk only comes from near-identical text, and preventing that is precisely what the practice reading and the paragraph-level measurement exist for. Richer clinician profiles make this stronger still: the more your team's real expertise and insight is on file, the more unmistakably yours every article becomes.