woods reads every public page, maps the estate, lists the claims it makes and shows where your content disagrees with itself. Each finding arrives with its evidence.
Public pages only. No sign-in, no credentials, nothing to install.The inferences may change as more is read. The observations do not.
Every environmental and energy claim on the pages woods reads, word for word, where it sits and when it changed. So you find the problem before a regulator, a competitor or a customer does.
How the green claims record worksEU: Directive (EU) 2024/825 applies from 27 Sep 2026 and bans generic environmental claims without recognised substantiation.
European Commission, Transition Pathwaysread 6 Oct 2026grade Aledger E2UK: the CMA has enforced consumer law directly since 6 Apr 2025, with fines up to 10% of global turnover. Green claims are an early focus.
Ashurst Perkins Coie briefingread 6 Oct 2026grade Cledger E1Showing the history of “100% renewable electricity”. Two pages changed their wording between 2 and 9 Sep 2026, and /help/business-faq still carries the old version. Which wording stands is your call.
| Claim | Type | Pages | First seen | Last seen | Change |
|---|---|---|---|---|---|
| Energy | 3 | 3 Feb 2026 | 14 Oct 2026 | 2 | |
Read 5 of 5 Pages disagree, 14 Oct 2026. Seen again, unchanged, on 2 pages. Cedar found a page that disagrees; which version stands is your call. This is the latest read. The claim, page by page
The claim’s history so far
Each read adds to the history. Nothing earlier is rewritten. | |||||
| Energy | 7 | 3 Feb 2026 | 14 Oct 2026 | 0 | |
| Environmental | 2 | 5 May 2026 | 2 Sep 2026 | 0 | |
| Environmental | 4 | 3 Feb 2026 | 14 Oct 2026 | 0 | |
| Environmental | 1 | 9 Sep 2026 | 14 Oct 2026 | 0 | |
Find yours. Each one opens the finding that answers it, from a real estate.
Each does one job, records what it saw and says what it could not see. woods keeps metadata, claims and where they sit, not the pages themselves.
Sweep complete: 318 pages, read by seven agents in run order. 38 need action. 22 were not read, and stay grey rather than score. Choose a band to see where it sits.
Maps every page, section and link, and states the distinct-page count with its duplicates shown.
Shows in Estate, Structure
Checks structure and machine-readability, finds orphans and dead ends, and groups faults by template.
Shows in Structure
Dates each page on evidence, names the evidence, and says “not known” when it cannot know.
Shows in Estate
Extracts every checkable claim in its exact words, with its page. The verdict stays with you.
Shows in Reports
Puts pages that disagree side by side, ranked by exposure. Which version is current is your call.
Shows in Reports
Scores each page’s voice against the estate’s own baseline, not a generic standard.
Shows in Reports
Renders each page and reports what a text crawler cannot see.
Shows in ReportsOn one real estate the map and the first findings landed in the first six minutes. The full sweep took about 106 minutes.
Who looks after each page. That needs read access to your own systems, and that connector is still being built. Until it ships, the estate map is the list to assign owners from.
Observations from one full crawl of the www host of a UK health brand, 31 Aug 2026, anonymised. Choose who is reading.
Showing the finding a Head of Content opens first: 182 HTML URLs, of which 131 are distinct pages on www. Read down from the reason to the evidence and its grade, then what woods could not see.
distinct pages on the www host. The crawl found 182 HTML URLs, 36 of them tracking variants of other pages.
Tracking variants, redirects and old sections inflate every count. A migration or a content plan built on 182 is built on the wrong number.
Own observation, anonymised. One estate, n = 1.
Pages that nothing links to and no sitemap lists. A crawl counts what it can reach, and woods states that limit rather than estimating past it.
The ledger line behind this figure does not break down the remaining 15 URLs, so this page does not either.
Plan from one true inventory. Retire the variants rather than counting them.
different descriptions of the company, each on a different page.
The positioning moved on and the older pages did not. Every reader, and every assistant that retrieves from the site, meets whichever version it lands on.
Own observation, anonymised. One estate, n = 1.
Which version is current. woods shows the versions side by side and ranks them by exposure. Choosing one is your call.
The three sentences are not quoted here, because quoting them would identify the brand. The report quotes each one beside its page.
Choose one story on purpose, then edit the pages still telling the others.
pages had no H1. 19 of those 26 sat on just two templates.
One template repeats one fault on every page it renders, so what looks like 26 separate problems is mostly two.
Own observation, anonymised. One estate, n = 1.
Why each heading is missing. The crawl records that it is absent, not who removed it or when.
Two template edits resolve 19 of the 26, which leaves 7.
pages said “clinically proven” in a title, a meta description or the H1.
Claims sit in titles, descriptions and body copy, written by many people over years. Nobody holds the list, so nobody can check it.
Own observation, anonymised. One estate, n = 1.
Whether any of the five is substantiated. woods lists each claim in its exact words, with its page and where on the page it sits. The verdict stays with you.
Work from the list: check each claim against its substantiation, then keep, edit or retire it.
HTML URLs sent a Last-Modified header.
A Last-Modified header is the plainest way a page tells a machine when it changed. Without one, an old page presents itself as current.
Own observation, anonymised. One estate, n = 1.
When any of these pages last changed, from this crawl alone. woods dates a page only on evidence: a change it observed between two reads, or a date the page or its sitemap declares. With neither, it says “not known” rather than guess.
Read the estate twice and the changes start to show. Until then, “not known” is the honest answer.
Four promises, each kept by a surface you will use in the app, by its real name.
The whole estate in one picture, by structure and health.
318 pages in 6 sections, each drawn at its share of the estate. In the app, opening a section lists the pages beneath it.
Nobody knows how many pages exist.
Tracking variants, redirects and old sections inflate every count.
Maps every page, section and link, and states the distinct pages with their duplicates shown.
One true inventory to plan from.
Faults look like hundreds of separate problems.
One template repeats one fault on every page it renders.
Finds orphans and dead ends, checks machine-readability, and groups faults by template.
Fix two templates and most of the fault is gone.
Every claim, every self-description, every voice, in the exact words.
06 · Sagethe fact-checkerComplete · 31m“I quote every claim in its own words, with its page. Whether it holds is yours to say.”
100% renewable electricity/tariffs/green-fix
Our eco-friendly heat pumps/heat-pumps
Carbon neutral since 2021/about
You cannot list what the site claims.
Claims sit in titles, descriptions and body copy, written by many people over years.
Extracts every checkable claim with its exact words and its page. The verdict stays with you.
A list legal can work from.
The brand says different things on different pages.
The positioning moved on. The older pages did not.
Shows the versions side by side, ranked by exposure. Which is current is your call.
One story, chosen on purpose.
Copy drifts as more people and tools write it.
There is no measured baseline to drift from.
Scores each page against the estate’s own baseline, not a generic standard.
Drift caught before readers notice it.
What aged, and how we know.
One tile opened from /help. Its band, and the evidence Rita used to date it, rung by rung.
Old pages present themselves as current.
Many sites declare no dates that a reader or a machine can check.
Dates each page on evidence, names the evidence, and says “not known” when it cannot know.
Decay found without guessing.
Keep, Edit or Retire, fixed at the template where one rule causes many faults.
Ordered by what it costs to leave alone. Colour is severity and nothing else.
WhyTwo template edits resolve 33 of the 41 pages with no H1. Page by page, it would take 33 edits.
WhySix pages say 2021 and two say 2022. If 2021 is right, two edits make the site agree with itself. Which year is right is your call.
WhyEach says something is green or eco-friendly with nothing on the page to back it. They are the claims most exposed if anyone asks what they rest on.
WhyEach declares a date over two years old and still presents itself as current to every reader and every assistant.
Your worklist: 1 to keep, 3 to edit, 0 to retire.
A report with no next step.
Findings arrive as scores, not decisions.
Keep, Edit or Retire for each page, and one fix for each rule that a template breaks.
A worklist, not a reading list.
Grade C is the lowest grade that publishes. Grade D never does. Each figure here comes from one estate, so each one says n = 1.
| Ledger | What was observed | Source | Date | Grade |
|---|---|---|---|---|
| E5 | 182 HTML URLs, 36 of them tracking variants, and 131 distinct pages on the www host. 26 of the 131 had no H1, and 19 of those sat on two templates. 0 of 182 sent a Last-Modified header. 5 pages said “clinically proven” in a title, description or H1. Three different self-descriptions. Readability across 130 pages ran from 60.0 to 90.9, median 75.5. | A full crawl of the www host of a UK health brand, with the counts recomputed by woods. | Crawled 31 Aug 2026, recomputed 3 Oct 2026 | AOwn observation, anonymised. One estate, n = 1. |
| E6 | On one real estate the map and the first findings landed in the first six minutes. The full sweep took about 106 minutes. | woods’ own engineering record of one sweep (design system lessons, item 20). | 2026 | AOne estate, n = 1. A larger or slower site may take longer. |
A: a primary source, directly verified. B: a primary source, verified through its official summary. C: a reputable secondary source.
Why the brand is not named: consent to name it is not on record, so every public page calls it a UK health brand.
Case file 01 (published 5 Oct 2026) reports a later woods sweep of the same brand, which read 745 pages. The two counts come from different tools, on different dates, so they are not expected to match. Every figure on this page is from ledger E5 or E6, and says which.
The Assessment reads your public estate and returns the evidence, in a form that survives the internal meeting. A number alone invites “says who?”, so every finding shows its working.
What you getRead-only. Public pages only. Your pages are read, not stored.
Assistants, site search and agents read what is published, current or not. woods shows you what they are reading, with the coverage, the method and an evidence grade on every finding.
Your estate, mapped and read, with the evidence beside every finding.