Noxia · The register
Every number on this site, where it came from, and when we last checked it.
Including the ones we took down. Almost every statistic published about AI in financial services is published by someone selling AI to financial services, and the only useful response to that is to show your working. This page is generated from the same data that renders the figures, so it cannot quietly disagree with them.
The short answer
Noxia publishes a source, a weight label and two dates beside every figure on the site: when it was last checked, and when it is next due. Volatile sources are re-checked every 30 days, annual datasets and rules in force every 180, dated research yearly. The build refuses to run once anything is more than 30 days past due. Figures we could not source were removed, and they are listed below rather than deleted quietly.
Figures we removed or corrected
This is the part of the register that matters. A list that only ever grows is a marketing page; one that records what came off it is a record.
| Figure | What it claimed | Why it went | When |
|---|---|---|---|
| 1,445% | a one-year surge in demand for multi-agent systems, attributed on our own homepage to Gartner | The attribution was false. The figure traces to a press release by Belitsoft, a software development company. No Gartner publication carrying it could be found. Borrowing an analyst’s name for a vendor’s number is the worst thing on this list, and it was ours. | 7 September 2026 |
| 25% | standard AI personalisation performing worse than sending nothing at all, across 500,000+ cold emails | No source found. The largest public dataset we could find — Hunter’s 31 million emails — reports the opposite: two custom attributes lift reply rate from 3.6% to 5.6%. A figure we could not source, contradicted by one we could. | 7 September 2026 |
| 127% | the same personalisation performing better than baseline after a deliverability rewrite pass | No source found. It appeared in our own plan document with no citation, which is how an internal guess becomes a public claim. | 7 September 2026 |
| 1.9× | one human running two AI seats booking more meetings per dollar than pure AI | No source found. The argument it was making survives — Hunter’s data shows manually edited emails beating fully automated ones by 18% — but that is a smaller, truer claim with 31 million emails behind it. | 7 September 2026 |
| 24/7 | set in the typography of a statistic beside three real figures | Not a statistic at all — a description of the offer, dressed as evidence. It belongs in the copy, and that is where it went. | 7 September 2026 |
| 95 / 85 / 70 / 65% | skill bars for Development, Design, Marketing and Consulting | Self-scored, unfalsifiable, and inherited from the WordPress template. At least they were honestly ours, which is more than the 1,445% managed. | 7 September 2026 |
| “low 80s” | the best computer-use score on OSWorld | Not removed — corrected. The real figure was 86.1%, against a 72.36% human baseline, so the sentence had quietly stopped making its own point: agents now beat people at bounded desktop tasks. Being out of date made the claim too modest, which is the failure mode nobody checks for. | 7 September 2026 |
| 1 in 5 · 40% · £39 | calls missed, bookings made after hours, cost per missed appointment — the specified defaults for our missed-call calculator | Every published figure we found came from a company selling call answering, and they disagree with each other by a factor of two. So the calculator shipped with no defaults at all, and asks you for your own numbers instead. | 8 September 2026 |
| 156% | pages combining text, images and structure being cited that much more often than text alone — the figure our thumbnails and diagrams were first justified with | No checkable method. We had it second-hand from an aggregator that itself marked it unverified; the origin, a December 2025 vendor blog, asserts it from 15,847 AI Overview results and never says how the multiplier was computed. The thumbnails stay, because they never needed a statistic: they are machine-readable and they are not stock photography. The number goes. | 11 September 2026 |
How the clock works
A checked date is a claim about the past, and a date nobody is obliged to renew is decoration. Each source carries a volatility instead, and volatility sets a cadence the build enforces.
| Volatility | What it means | Re-checked |
|---|---|---|
| live | A leaderboard, an index, a price. Moves without notice and without a version number. | every 30 days |
| dataset | An annual industry dataset. Will not change — it will be superseded, which is the same problem later. | every 180 days |
| regulatory | A rule in force. Rarely changes; catastrophic when it does and nobody noticed. | every 180 days |
| dated | A dated publication whose figure is fixed. Re-read yearly to confirm it has not been retracted or restated. | yearly |
| fixed | Arithmetic, our own stated opinion, or a historical artifact quoted as one. Saying “never” is more honest than a date nobody will act on. | never |
The finding behind this: pages cited by AI search skew about 25% fresher than the organic top ten. The tempting reading is “publish more often”. The useful one is narrower — a page carrying a checkable number is only citable while the number is still true, and the figure we removed above was not old, it was wrong. A cadence catches both.
Every source currently on the site
74 of them, across 28 pages.
| Page | Source | Weight | Checked | Next due |
|---|---|---|---|---|
| Half of what brokers send lenders is wrong first time. Here is where the hours go. | Nivo, research with 70+ specialist lenders and brokers, reported by Mortgage Solutions, 7 May 2026. Figures: 50% right first time, 15–20 message rounds, 5+ people, 5+ hours broker-to-lender admin, 35% corrections, 20–50% stalling. | vendor research, named method and sample | 2026-09-08 | 2027-03-07 |
| Can an IFA use AI to write suitability reports without breaching Consumer Duty? | AdvisoryAI, AI Suitability Report Generators for UK Financial Advisers: 2026 Buyer’s Guide, 18 May 2026 — the four-to-six-hours-per-report figure. | vendor selling the fix; uncorroborated | 2026-09-08 | 2027-03-07 |
| Can an IFA use AI to write suitability reports without breaching Consumer Duty? | FCA Consumer Duty, in force 31 July 2023 — the four outcome tests that make this a records question. | regulator, primary | 2026-09-08 | 2027-03-07 |
| AI agents now beat humans at desktop tasks. They still finish 30% of real work. | OSWorld-Verified leaderboard — 86.1% (Qwen3.8-Max), top five 83.4–86.1%, against the 72.36% human baseline stated in the original OSWorld paper for its 369 tasks. Leaderboard data verified 4 September 2026; re-checked by us 8 September 2026. | public leaderboard, primary | 2026-09-08 | 2026-10-08 |
| AI agents now beat humans at desktop tasks. They still finish 30% of real work. | TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks, Carnegie Mellon, arXiv:2412.14161 — "the most competitive agent can complete 30% of tasks autonomously". | peer-reviewable preprint, primary | 2026-09-08 | — |
| Do the arithmetic on your unanswered phone before you buy anything. | No external source is used for the figures in this note. The illustration is explicitly illustrative, and the method is arithmetic on your own call log and CRM. | no borrowed statistics, on purpose | 2026-09-08 | — |
| Per-minute AI pricing charges you for the thing working. | No external statistics are used in this note. The argument is about contract shape, and the illustration is explicitly illustrative. | argument, not evidence — labelled as such | 2026-09-08 | — |
| In a regulated firm, the audit trail is the product. | FCA Consumer Duty (in force 31 July 2023) and SM&CR — the accountability requirements this specification is written against. | regulator, primary | 2026-09-08 | 2027-03-07 |
| In a regulated firm, the audit trail is the product. | The six-record specification is Noxia’s own and is not a regulatory standard. It is what we build to. | our opinion, labelled | 2026-09-08 | — |
| We checked five AI statistics we were about to publish. Three had no source. | Gartner, Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025, 26 August 2025. | analyst, primary | 2026-09-08 | 2027-09-08 |
| We checked five AI statistics we were about to publish. Three had no source. | OSWorld-Verified leaderboard — 86.1% best agent (Qwen3.8-Max); 72.36% human baseline from the original OSWorld paper. Re-checked 8 September 2026, still the top score. | public leaderboard, primary | 2026-09-08 | 2026-10-08 |
| We checked five AI statistics we were about to publish. Three had no source. | TheAgentCompany, Carnegie Mellon, arXiv:2412.14161 — 30% of tasks completed autonomously. | preprint, primary | 2026-09-08 | — |
| We checked five AI statistics we were about to publish. Three had no source. | Hunter, State of Email Outreach 2026 — 31 million emails; two custom attributes 5.6% vs 3.6% reply; manually edited 5.2% vs fully automated 4.4%. | vendor dataset, large, method stated | 2026-09-08 | 2027-03-07 |
| We checked five AI statistics we were about to publish. Three had no source. | Belitsoft press release, Multi-Agent Systems Surge 1,445%… — the actual origin of the figure we had credited to Gartner. | vendor press release — the point of the story | 2026-09-08 | — |
| What to automate first in an advice firm — and what to never automate. | OSWorld-Verified leaderboard (86.1%, best agent) and TheAgentCompany, Carnegie Mellon, arXiv:2412.14161 (30% autonomous completion). Re-checked 8 September 2026. | primary | 2026-09-08 | 2026-10-08 |
| What to automate first in an advice firm — and what to never automate. | The two-axis framework is Noxia’s own and is not a regulatory classification. | our opinion, labelled | 2026-09-08 | — |
| Does AI-written outreach work? 31 million emails say something uncomfortable. | Hunter, State of Email Outreach 2026 — 31 million emails sent by Hunter users in 2025. Two custom attributes 5.6% vs 3.6%; manually edited 5.2% vs fully automated 4.4%; 69% of decision makers bothered by AI-written emails. | vendor dataset, large, method and sample stated | 2026-09-08 | 2027-03-07 |
| Your firm may be invisible to ChatGPT. Here is how to check in ten minutes. | Cyrus Shepard, AI Citation Ranking Factors, Zyppy Signal, 7 May 2026 — 23 factors scored from 54 experiments, patents and case studies, weighted by repeatability, strength of evidence and platform documentation: URL accessibility 9.5/10, search rank 9.4, fan-out rank 9.3, preview controls 9.2, query-answer match 9.2. The author states the scores reflect correlation, not causation. | third-party synthesis of 54 sources, method stated; not a platform disclosure | 2026-09-11 | 2027-03-10 |
| Your firm may be invisible to ChatGPT. Here is how to check in ten minutes. | Ryan Law, Do AI assistants prefer to cite fresh content?, Ahrefs, 28 July 2025 — 16.975 million cited URLs across ChatGPT, Perplexity, Gemini, Copilot, AI Overviews and organic Google: cited content averaged 1,064 days since publication against 1,432 for the organic top-10, 25.7% fresher. | vendor dataset, very large, method stated | 2026-09-11 | 2027-03-10 |
| Your firm may be invisible to ChatGPT. Here is how to check in ten minutes. | BuzzStream, 50 AI statistics digital PR and SEO should know, updated 30 June 2026 — across 4 million citations tracked with Xofu, blog and content pages made up 53.46% of citations; syndicated press releases 0.04%. | vendor dataset (a digital-PR firm), large; sells the conclusion | 2026-09-11 | 2027-03-10 |
| The cheapest fix for missed appointments is one line in a text message. | Hallsworth M et al., Stating Appointment Costs in SMS Reminders Reduces Missed Hospital Appointments: Findings from Two Randomised Controlled Trials, PLOS ONE, 10.1371/journal.pone.0137306. Trial one Nov 2013 – Jan 2014, n = 10,111: 11.1% vs 8.4% DNA, OR 0.74 (95% CI 0.61–0.89, p<0.01). Trial two Mar – May 2014, n = 9,848: specific 8.2% vs general 9.9%, OR 1.22 (95% CI 1.00–1.48, p = 0.046). Barts Health NHS Trust, five outpatient specialties. | peer-reviewed randomised controlled trials, primary | 2026-09-08 | — |
| The cheapest fix for missed appointments is one line in a text message. | The three-step sequence — wording first, arithmetic second, automation third — is Noxia’s own and is not part of the study. | our opinion, labelled | 2026-09-08 | — |
| Nobody answers the phone after six, and your customers already know. | InStep AI, reported by Property Industry Eye — 94.1% of calls to 3,047 unique UK estate and letting agency numbers unanswered between 6pm and 9pm, calls placed May–July 2026. The publisher sells “Ella”, an out-of-hours AI answering assistant, and disclosed both its limitations and its commercial interest in the report. | vendor research; method, sample and commercial interest all stated | 2026-09-08 | 2027-03-07 |
| Nobody answers the phone after six, and your customers already know. | The figures we declined: published claims for unanswered UK business calls generally, ranging from roughly half to nearly all, none with a stated method. Listed on the public register rather than used. | not published — recorded as rejected | 2026-09-08 | — |
| Twenty-six machinery standards stop counting on the same day. | GOV.UK designated standards — all thirteen consolidated lists, read through the GOV.UK content API and parsed from the published OpenDocument spreadsheets on 8 September 2026. 3,517 standards, 699 with a cessation date, 634 of those passed and 65 still ahead. Every count in this note is from that reading and can be checked against the same lists. | primary; the department that maintains the lists | 2026-09-08 | 2027-03-07 |
| Twenty-six machinery standards stop counting on the same day. | The January 2027 cluster: 30 of the 65 remaining deadlines fall in that month, 26 of them on 20 January and all of those on the machinery list. Counted from the same reading. | our own count from the primary source | 2026-09-08 | 2027-03-07 |
| Twenty-six machinery standards stop counting on the same day. | How much NOTICE a cessation date gives could not be measured and is not claimed anywhere in this note. A single reading shows that a date exists, not when it was published. Establishing that needs two readings, which is why the corpus is kept rather than re-fetched. | a limitation, stated rather than glossed | 2026-09-08 | — |
| Twenty-six machinery standards stop counting on the same day. | No standard’s text is reproduced here. ISO, IEC and BSI documents are licensed publications; this note names references and reports what the government list says about them. | a limitation we chose | 2026-09-08 | — |
| One in three UK businesses now use AI. One in ten use it seriously. | Office for National Statistics, “Artificial intelligence in UK businesses: 2023 to 2026”, published 20 July 2026. Business Insights and Conditions Survey wave 159, fieldwork 15 to 28 June 2026, 38,637 responses, response rate 26.7%. Every business figure in this note — 35%, 12%, 28%, 49%, 58%, 13%, 1.4 to 1.6, 10%, 18%, 16%, 12%, 6%, 2%, 41%, 18%, 11% — is from this release. | primary; official statistics from the national statistics office | 2026-09-11 | 2027-03-10 |
| One in three UK businesses now use AI. One in ten use it seriously. | The 55% employee figure is from the ONS Opinions and Lifestyle Survey for May to June 2026, as cited in the same release. It is a household survey with a different question and includes use for education, so the twenty-point gap against the business figure is indicative rather than exact. | primary, but a different survey from the business figure it is set against | 2026-09-11 | 2027-03-10 |
| One in three UK businesses now use AI. One in ten use it seriously. | All figures are self-reported by businesses. The ONS does not audit whether a reported AI technology is in production use; “extensively” is the respondent’s own word. We have not adjusted anything. | a limitation of the source, stated | 2026-09-11 | — |
| 18 of 45 citations did not exist. What a court said about AI drafting. | Ayinde, R (on the application of) v The London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC and another [2025] EWHC 1383 (Admin), Divisional Court (the President of the King’s Bench Division and Mr Justice Johnson), 6 June 2025. Every quotation and figure in this note — five fictitious cases (paragraphs 36 and 37), forty-five citations of which eighteen do not exist (paragraph 74), the statements at paragraphs 6, 7, 9 and 23, the £2,000 wasted costs orders (paragraph 49), and the referrals (paragraphs 70, 72 and 81) — is from the published judgment, read in full. | primary; the judgment itself | 2026-09-11 | — |
| 18 of 45 citations did not exist. What a court said about AI drafting. | The application to financial advice and other regulated work is our reading, not the court’s. The judgment concerns the conduct of litigation and the duties of lawyers; we have said where the analogy is ours. | our own argument, labelled as such | 2026-09-11 | — |
| Seven million people have until November to prove who they are. | Companies House, “Companies House confirms identity verification rollout from 18 November 2025”, press release, 5 August 2025: “We estimate that 6 to 7 million individuals will need to verify their identity by mid-November 2026”; “More than 300,000 individuals have already done so during the current voluntary period, which started in April”; existing directors confirm at their next confirmation statement during a 12-month transition; “It will be an offence to act as a director without being verified once director duties commence.” | primary; the department that runs the register | 2026-09-11 | 2027-03-10 |
| Seven million people have until November to prove who they are. | Companies House, “One million people verify identity early ahead of Companies House changes”, press release, 3 November 2025: “More than 1 million individuals have already verified their identity since we rolled out our identity verification service on a voluntary basis in April.” | primary | 2026-09-11 | 2027-09-11 |
| Seven million people have until November to prove who they are. | Companies House management information: identity verification, April to June 2026, published 30 July 2026 — quarterly data on the number and percentage of director and PSC appointments that comply. Published as ODS and XLSX only, and not opened by us — the figures in this note attributed to it are PKF Francis Clark’s reading, labelled as theirs. It is named here so you can open it yourself, and so that the next time this note is checked the first job is to replace their reading with ours. | primary, not opened by us — named so you can | 2026-09-11 | 2026-10-11 |
| Seven million people have until November to prove who they are. | PKF Francis Clark, “Companies House identity verification: enforcement is getting closer”, updated 11 August 2026, reading the April to June 2026 release as 55% of directors, 50% of LLP members and 42% of PSCs verified, and quoting Companies House’s stated intention to take proportionate action against those who fail to verify. | secondary; an accountancy firm’s reading of the primary release above. Quoted as theirs, not confirmed by us | 2026-09-11 | 2026-10-11 |
| You may not need the UKCA mark at all. What the CE decision covers. | GOV.UK, “Using the UKCA marking”, guidance last updated 21 August 2026: “Where the aims and requirements of product regulations meet the needs of the UK, the UK continues to recognise the CE (European Conformity) marking, alongside or in place of the UKCA (UK Conformity Assessed) marking, for the Great Britain market under The Product Safety and Metrology (Amendment) Regulations 2024.” The seven excluded areas are listed on the same page. | primary; the department’s own guidance | 2026-09-11 | 2027-03-10 |
| You may not need the UKCA mark at all. What the CE decision covers. | The Product Safety and Metrology etc. (Amendment) Regulations 2024, SI 2024/696. The explanatory note describes provision “allowing relevant economic operators to meet any conformity assessment or testing requirements in the legislation as it applies in Great Britain by complying with the corresponding EU conformity assessment or testing requirements.” The count of twenty-one amended instruments is ours, from the contents. | primary; the legislation | 2026-09-11 | 2027-03-10 |
| You may not need the UKCA mark at all. What the CE decision covers. | The divergence between designated and harmonised standards, and the 699-of-3,517 figure, are from our own reading of the thirteen GOV.UK consolidated lists on 8 September 2026. | our own count from a primary source | 2026-09-11 | 2027-03-10 |
| You may not need the UKCA mark at all. What the CE decision covers. | None of this is a conformity assessment or legal advice about a particular product. It is what two government pages and one statutory instrument say, read on the dates given. | a limitation, stated | 2026-09-11 | — |
| Four months into the Renters’ Rights Act: the clocks an agent keeps. | GOV.UK, “When will the Renters’ Rights Act come into force?”, news story published 30 April 2026: “On 1 May 2026, new laws kicked in” for 11 million renters; the database “will be rolled out gradually by area from late 2026”; the ombudsman is part of phase two “from late 2026 onwards”. | primary; the department | 2026-09-11 | 2027-03-10 |
| Four months into the Renters’ Rights Act: the clocks an agent keeps. | GOV.UK, “Guide to the Renters’ Rights Act”, updated 6 November 2025: Royal Assent 27 October 2025; existing fixed terms convert to periodic tenancies; rent increases once a year by section 13 notice with at least two months’ notice; asking rent must be published and offers above it are illegal to accept; advance rent capped at one month or 28 days; pets cannot be unreasonably refused; sale and move-in grounds need four months’ notice and cannot be used in the first twelve months; civil penalties up to £7,000 and up to £40,000 or prosecution; ombudsman membership required for all private landlords in England with assured or regulated tenancies; letting or advertising an unregistered property is enforceable. | primary; the department’s own guide | 2026-09-11 | 2027-03-10 |
| Four months into the Renters’ Rights Act: the clocks an agent keeps. | The framing of the Act as a set of clocks, and the observation that the noticing rather than the judging is what fails at scale, is ours. The Act applies to England; Scotland, Wales and Northern Ireland have their own tenancy law. | our own argument, and a scope limitation, stated | 2026-09-11 | — |
| 26% of UK adults trust ChatGPT for financial advice. The FCA counted. | Financial Conduct Authority, “AI and the future of retail financial services (The Mills Review)”, 6 July 2026. Executive summary: “Around 26% trust general-purpose tools such as ChatGPT, Claude or Gemini for financial advice, despite limited awareness that formal routes to recourse will not apply”; “1 in 5 UK adults are already open to AI making decisions for them, with demand strongest where choices feel complex or high-stakes, particularly debt advice, pensions and investments.” Consumer research by Yonder Consulting, 5,026 UK adults, April 2026. Seven recommendations, the first “Secure and adapt the regulatory perimeter”. | primary; the regulator’s own review and survey | 2026-09-11 | 2027-09-11 |
| 26% of UK adults trust ChatGPT for financial advice. The FCA counted. | FCA press release, “FCA publishes landmark review into impact of AI on retail financial services”, 6 July 2026: “a fifth of people – equivalent to 11 million UK adults – are likely to use AI that can act autonomously within pre-set goals”; over 5,000 consumers surveyed in April 2026. | primary | 2026-09-11 | 2027-09-11 |
| 26% of UK adults trust ChatGPT for financial advice. The FCA counted. | Bank of England and Financial Conduct Authority, “Artificial intelligence in UK financial services – 2024”, 21 November 2024. 118 firms: 75% using AI, 10% planning to within three years; a third of use cases third-party implementations; top three providers 73% of cloud, 44% of models, 33% of data; 55% of use cases with some automated decision-making, 2% fully autonomous; 46% of firms with partial understanding against 34% complete; foundation models 17% of use cases. | primary; the regulators’ own survey. A fourth edition is being run in 2026 and will supersede these figures when it is published | 2026-09-11 | 2027-03-10 |
| 26% of UK adults trust ChatGPT for financial advice. The FCA counted. | House of Commons Treasury Committee, “Artificial intelligence in financial services”, Fifteenth Report of Session 2024–26, 20 January 2026: “Some 75% of UK financial services firms are now using AI”; the FCA asked to publish comprehensive guidance, including SM&CR accountability, by the end of 2026; HM Treasury asked to designate major AI and cloud providers as critical third parties by the end of 2026; no firm had been brought into the regime as of October 2025. | primary; a select committee’s recommendations, which bind nobody until acted on | 2026-09-11 | 2027-09-11 |
| 26% of UK adults trust ChatGPT for financial advice. The FCA counted. | The reading of the recourse gap as an adviser’s argument is ours. The review does not tell advisers what to say to clients; it reports what clients are doing. | our own argument, labelled as such | 2026-09-11 | — |
| Financial services | Nivo, research with 70+ UK specialist lenders and brokers, reported by Mortgage Solutions, 7 May 2026 — 50% right first time, 15–20 message rounds, 5+ people, 5+ hours broker-to-lender admin, 35% corrections. | vendor research, named method and sample | 2026-09-08 | 2027-03-07 |
| Financial services | OSWorld-Verified leaderboard — 86.1% best agent (Qwen3.8-Max) across 369 tasks, against the 72.36% human baseline stated in the original OSWorld paper. Leaderboard data verified 4 September 2026; re-checked by us 8 September 2026. | public leaderboard, primary | 2026-09-08 | 2026-10-08 |
| Financial services | TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks, Carnegie Mellon, arXiv:2412.14161 — "the most competitive agent can complete 30% of tasks autonomously". | peer-reviewable preprint, primary | 2026-09-08 | — |
| Financial services | FCA Consumer Duty, in force 31 July 2023, and SM&CR — the accountability requirements the six-record specification is written against. | regulator, primary | 2026-09-08 | 2027-03-07 |
| Financial services | The six-record specification and the four-item exclusion list are Noxia’s own. Neither is a regulatory standard. They are what we build to. | our opinion, labelled | 2026-09-08 | — |
| Lettings | Goodlord, State of the Lettings Industry — 2,750+ respondents including 350+ letting agents, 1,350+ landlords and around 1,000 tenants. Typical agency 340 managed properties with 14 employees; 54% employ five or fewer; 47% experimenting with AI; about 27% of agency revenue from tenancy renewals. | vendor research; sample stated, full method not disclosed | 2026-09-08 | 2027-03-07 |
| Lettings | InStep AI, reported by Property Industry Eye — 94.1% of calls to 3,047 unique UK estate and letting agency numbers unanswered between 6pm and 9pm, calls placed May–July 2026. Publisher sells an out-of-hours AI answering assistant and disclosed its commercial interest. | vendor research; method, sample and interest all stated | 2026-09-08 | 2027-03-07 |
| Lettings | Hallsworth M et al., Stating Appointment Costs in SMS Reminders Reduces Missed Hospital Appointments, PLOS ONE 10.1371/journal.pone.0137306 — 11.1% to 8.4% DNA across 10,111 patients. Cited here for the mechanism, not the setting. | peer-reviewed randomised controlled trial, primary | 2026-09-08 | — |
| Lettings | The Renters’ Rights Act’s effect on renewal fees, and the six-record specification, are Noxia’s reading and Noxia’s own standard respectively. Neither is legal advice. | our opinion, labelled | 2026-09-08 | 2027-03-07 |
| What is your unanswered phone actually costing you? | No industry statistic is used to compute anything on this page. The arithmetic runs on numbers you supply from your own call log and CRM. | no borrowed statistics, on purpose | 2026-09-08 | — |
| What is your unanswered phone actually costing you? | InStep AI, reported by Property Industry Eye — 94.1% of calls to 3,047 UK estate and letting agency numbers unanswered between 6pm and 9pm, calls placed May–July 2026. Quoted here only as evidence that the published range is wide and vendor-supplied; not used as a default. | vendor research; commercial interest disclosed by the publisher | 2026-09-08 | 2027-03-07 |
| Should you automate this? Six questions that settle it. | The two-axis framework (reversible × detectable) and the four supporting questions are Noxia’s own. They are not a standard, a regulation or a published methodology. | our opinion, labelled | 2026-09-08 | — |
| Should you automate this? Six questions that settle it. | The reliability premise — that bounded tasks succeed and long-horizon cross-application work does not — rests on the OSWorld-Verified leaderboard (86.1% best agent, 72.36% human baseline) and TheAgentCompany, Carnegie Mellon, arXiv:2412.14161 (30% completed autonomously). | public leaderboard and preprint, both primary | 2026-09-08 | 2026-10-08 |
| Should you automate this? Six questions that settle it. | No answers, scores or results leave the page. The page counts its own views, anonymously and first-party, and nothing you click is part of that count. | verifiable in your browser’s network tab | 2026-09-08 | — |
| Which UK product rules apply to what you make? | GOV.UK designated standards — thirteen consolidated lists, 3,517 standards, ingested into notes/standards.json and rebuilt by build-standards.mjs. Each result links back to the list it came from, where the position is authoritative. |
government publication, primary | 2026-09-08 | 2027-03-07 |
| Which UK product rules apply to what you make? | The six questions and the mapping from answer to regime are Noxia’s own. They are a navigation aid, not a legal test, and no part of this page is a conformity assessment. | our opinion, labelled | 2026-09-08 | — |
| Can an AI assistant actually read your website? | Cyrus Shepard, AI Citation Ranking Factors, Zyppy Signal, 7 May 2026 — URL accessibility 9.5/10, search rank 9.4, fan-out rank 9.3, preview controls 9.2, query-answer match 9.2, scored from 54 sources and stated by the author to be correlations. The ordering of these checks follows those weightings. | third-party synthesis, method stated; not a platform disclosure | 2026-09-11 | 2027-03-10 |
| Can an AI assistant actually read your website? | The six checks, their wording and the fixes are Noxia’s own. No URL is fetched and nothing you enter leaves the page. | our opinion, labelled | 2026-09-11 | — |
| Is this supplier in trouble? Where to look, free. | The Gazette (thegazette.co.uk) — the UK’s official public record, carrying statutory insolvency notices. Free to search, no account. | official public record, primary | 2026-09-08 | 2027-03-07 |
| Is this supplier in trouble? Where to look, free. | Companies House (find-and-update.company-information.service.gov.uk) — filing history, accounts due dates, charges register, officers, and the register of disqualifications. Free to search, no account. | government register, primary | 2026-09-08 | 2027-03-07 |
| Is this supplier in trouble? Where to look, free. | The mapping from exposure to signal, and the ordering, are Noxia’s own. Nothing here is credit advice, and no supplier name is typed, stored or transmitted by this page. | our opinion, labelled | 2026-09-08 | — |
| What can the things you installed actually reach? | The six categories and the blast-radius ordering are Noxia’s own, drawn from building Agent Market Store. They are not a standard, a framework or a certification. | our opinion, labelled | 2026-09-08 | — |
| What can the things you installed actually reach? | Nothing is scanned, no system is named, and no answer leaves the page. The page counts its own views, anonymously and first-party, and nothing you answer is part of that count. | verifiable in your browser’s network tab | 2026-09-08 | — |
| What is the going day rate? Not from two data points, it isn’t. | No rate data is published by this page. Every figure in the result is one you entered, and the arithmetic is a median and a range over your own points. | no borrowed statistics, on purpose | 2026-09-08 | — |
| What is the going day rate? Not from two data points, it isn’t. | The n≥3 floor, the 60% spread warning and the source-attribution rule are Noxia’s own. DayRate, the rate index this was specified against, has no repository — recorded on the work page rather than implied. | our opinion, labelled | 2026-09-08 | — |
Questions people actually ask
Where does Noxia get its statistics?
Every number on the site is listed on this page with its primary source, a label saying how much weight it carries, the date it was last checked and the date it is next due. Where a source is a vendor selling the answer, the label says so. Where a claim is our own opinion rather than evidence, the label says that too.
What happens when a figure goes out of date?
Each source carries a re-check cadence set by how volatile it is: leaderboards and live indices every 30 days, annual datasets and rules in force every 180, dated research yearly, arithmetic and stated opinions never. The site build reports what is due and refuses to build once any source is more than 30 days past due, so a stale figure stops a deploy rather than sitting there.
Has Noxia ever published a figure that was wrong?
Yes, and they are listed on this page. The worst was a 1,445% figure on our own homepage credited to Gartner; it traces to a press release by a software development company and no Gartner publication carrying it could be found. It was removed on 7 September 2026 along with three unsourced outbound-marketing figures and a set of self-scored skill bars.
Why publish the numbers you got wrong?
Because the alternative is asking people to take an agency’s numbers on trust in a market where almost every published statistic is produced by someone selling the fix. A register that only ever grows is a marketing page. One that records what came off it is a record.
Found a number here that has gone stale?
Tell us and we will correct it and say what changed. That is not a courtesy — it is the only thing that makes the rest of this page worth anything.
Tell us what is wrong