Our own site · UK fraud markers and de-banking · 17 November 2026 · By Chris Aylott

FraudMarkers: what our audit changed, and what we refused to claim

FraudMarkers is our own site.I run it, and we used it to test Chad’s audit end to end on a YMYL topic, from finding to shipped fix. It has two jobs, and we want to be plain about both. It is a test site: we use it to learn what works for legal information in a YMYL niche (what gets read, found and cited). And it is built, in its own right, to help people deal with fraud markers and de-banking, as an independent guide where every statement is sourced.

This is what we checked, what we changed, what we refused to claim, and what we will measure next. We have not measured any results yet, so there are none here.

Largest deduction
Fixed
−15, local business type with no address
New decision pages
4
one sourced table each
Quotes verbatim
194 / 218
every miss reviewed by hand
AI results claimed
0
tracking starts next

The site

FraudMarkersis a UK guide to CIFAS fraud markers, frozen and closed bank accounts and de-banking. Getting it wrong costs readers money and access to banking, which is exactly the “your money or your life” territory where a confident, unsourced answer does harm. It also lists solicitors who handle these cases. The directory is alphabetical; no firm pays to appear, and nothing ranks one firm above another.

The plumbing was already in good shape for machines, and we checked each piece on the live site on 17 November:

  • A static Astro build, so the content is in the HTML the server sends, not assembled by JavaScript.
  • A Markdown twin of every page: all 50 URLs in the sitemap advertise one with <link rel="alternate" type="text/markdown">, and all 50 return it as text/markdown.
  • An /llms.txt that summarises each guide in its own words.
  • A robots.txt that allows every crawler, and pages that return 200 to the GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot user agents.

The finding we fixed first

Chad’s quick scan scored the site 81. The largest single deduction (−15) came from the directory: every crawled firm page, 13 of them, declared the firm as LegalService, a local business type, with no address. That failed local/business-identity, the rule that asks whether a machine can identify the place a page says it is.

The fix was not to add addresses. A local business type says “this page is the business”, and a directory is not the firm: it describes firms and does not speak for them. So each firm is now a plain Organization, keeping its name, website, phone, SRA number, area served and areas of practice. On the built pages, the local rules are now not applicable on all 14 firm pages instead of failing.

What the scan could not see

The same scan read 25 pages, and the thin directory pages (about 100 words each) took 13 of them, so it assessed only two guides. A quick scan tells you about the pages it read. We logged the thin pages as their own task rather than read the score as a verdict on the whole site.

Four decision pages, every claim checked

Chad’s content/decision-pages finding (are there pages that help someone choose between options?) and its content/information-gainfinding (three guides cited only third-party sources) shaped the next four guides. Each compares options in one sourced table, shows a visible “Last reviewed” date and ends with an FAQ. None names or ranks a firm.

GuideWhat it comparesClaims checkedChanged by the fact-check
Frozen vs closed bank accountA bank freeze, a closure and a court freezing orderabout 346 fixed, 2 removed
Does a CIFAS marker affect your credit score?A CIFAS marker, a CCJ and a defaultabout 406 fixed
SIRA, National Hunter and CIFAS markersThree fraud prevention databases627 fixed
Challenging a CIFAS markerThe member, Cifas, the Ombudsman and the ICO6114 fixed

The guide to challenging a CIFAS marker shows the shape: four routes side by side, with what each body can and can’t do, what you need first, the time limit, the cost and the published timescale. Where no source states something, such as a fee, the table says no fee is stated rather than calling it free.

How the research was checked

The research started as deep-research fact sheets from Parallel. We treated those as leads, not sources. Every quoted passage was checked against the full text of the page it cited:

Result of the checkPassages
Verbatim on the cited page194
Near match (85% or more of five-word runs)5
Partial match11
Not found on the cited page6
No source URL given2

We reviewed the misses by hand. One was a paraphrase in quotation marks: the fact sheet had HM Treasury’s basic bank account page saying banks are “legally required to offer” one, while the page says HM Treasury “has designated the nine largest personal current account (PCA) providers in the UK to offer basic bank accounts”. Most of the rest were formatting: a bullet list flattened into a sentence, clause numbering, links inside the text. A few could not be found on the cited page at all, among them retention periods for SIRA and National Hunter, so the guide says the operator does not publish them.

A fact-check pass then went through each guide claim by claim. It mostly caught overstatement, not invention: “15 days” became “15 business days” for payment services complaints, the Ombudsman’s timescales became service-wide targets rather than results, and “only the filing organisation can remove it” lost the “only”. It also found that the fact sheet was out of date: National Hunter is now run by Experian.

Recording dated public facts about listed firms

The directory now shows three dated, public facts about each firm, so readers can check them:

  • its Google rating, with the date it was checked and a link to the profile. It is never used to order the list;
  • the year the firm says it was founded, linked to the page where it says so;
  • its Companies House details, only when the firm’s own website gives the company number.

Each value goes into an append-only register, dated, with a reason recorded wherever it is unknown. That is for accuracy, and so that changes show up. One did straight away. The first Companies House pass found that Rahman Ravelli Solicitors Ltd is in administration: administrators were appointed on 20 October 2026. We took the listing down, because the directory lists firms that can take on new cases, and published a sourced timeline of the administration for its clients. No reason for the administration has been published, and the page says so. It is not a judgement on the firm’s work, and the SRA register shows no disciplinary or regulatory decisions published about it.

What we deliberately did not do

No rating markup. The Google ratings are on the page but not in aggregateRating. Google’s review snippet guidelines say “Don’t aggregate reviews or ratings from other websites”, and these come from Google, not from FraudMarkers’ readers. The firm’s schema links the profile in sameAs instead.

No claim that schema gets you cited. The schema fix makes the markup true. We are not claiming it will make AI assistants cite the site. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 matched control pages, and found no meaningful uplift in citations in Google AI Overviews, AI Mode or ChatGPT (published 11 May 2026). It is a matched-control study, not a randomised trial, so it doesn’t prove schema can never matter. It is the best evidence we have, and it doesn’t support the claim.

It also applied to us. Four of Chad’s own schema rules told customers that structured data gets a page cited or quoted by AI. On 17 November we reworded all four, and added a build gate (Gate H) that fails if a schema rule’s wording makes that promise again. It is on our changelog.

What we will measure

  • AI visibility. Mentions and citations of FraudMarkers in ChatGPT, Claude, Gemini and Perplexity, for the questions these guides answer. Tracking starts next.
  • A Financial Ombudsman decisions tracker. Published fraud-marker decisions, counted and dated, as data of the site’s own that its sources do not already hold.

We will publish the results, whichever way they go, in a follow-up. Until then, nothing on this page should be read as evidence that any of these changes worked.

If you run content for a real site

  • Read a schema finding as a question about what your page claims to be. Sometimes the fix is a less specific type, not more fields.
  • Check how many of your pages a scan read before you trust its score.
  • Treat research tools as leads. Check every quotation against the page it cites; the errors that survive are overstatements, and they read as fine.
  • Show third-party ratings, dated and linked, without marking them up as your own.
  • Don’t promise an AI-citation effect you have not measured, including in your own tools.

Sources

  • TryChad.ai: Chad quick scan of fraudmarkers.com, 17 November 2026 (score 81; local/business-identity on 13 firm pages, −15). Primary source.
  • TryChad.ai: chad-audit changelog, 17 November 2026: four schema rules reworded; Gate H. Primary source.
  • FraudMarkers: the schema fix, the four guides with their fact-check tallies, the passage check (218 passages) and the directory register, all 17 November 2026. Primary source; we run the site.
  • FraudMarkers: Challenging a CIFAS marker and Rahman Ravelli administration timeline.
  • Ahrefs: We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved, 11 May 2026. ahrefs.com
  • Google Search Central: Review snippet structured data guidelines. developers.google.com

More client work is in the case studies.

How does your site look to Chad?

Drop your domain. In 11 seconds you get your GEO score, your biggest gap, and the first thing Chad would fix.

Free scan, usually under a minute. No sign-up.