In 2023, audited energy performance certificates accounted for 0.08% of all residential certificates issued in France that year. The figure comes from the French Court of Auditors, in its June 2025 report on the implementation of the DPE — the Diagnostic de Performance Énergétique, France's energy performance certificate. Across the Channel, in England and Wales, the rules require accreditation bodies to audit at least 2% of lodged certificates every year. A ratio of 1 to 25.
That number is going to circulate, and it is going to be misread. Because the two countries do not audit the same object: France supervises professionals, the UK audits certificates. The gap measures a difference in architecture first — and says nothing, either way, about the quality of the certificates produced.
The stakes are concrete. A label determines the right to let a home, now that the least efficient properties are being phased out of the rental market; it unlocks renovation grants; and it moves prices. According to notarial data cited by the Court of Auditors, a house rated F or G sells at a discount of 5% to 22% against an equivalent house rated D, depending on the region. On a €250,000 property, one letter can be worth more than the renovation itself.
0.08% of French EPCs were audited in 2023, against a regulatory minimum of 2% of certificates in England and Wales. But France audits assessors where the British regime audits assessments — and since October 2025 it screens 100% of its national database statistically, something almost no other European country does.
What this article covers
We set out the British regime, often held up as a model but rarely described; we compare eight other European systems, from the strictest to the most discreet; we explain what the 2024 European directive has required since 29 May 2026 and why it changes the very nature of the obligation; we put a number on what stronger auditing would cost in France; and we show why comparing "error rates" across countries is the main trap in this subject. The scope is existing homes in mainland France, and certificates — not energy audits or renovation grants.
The framework: what Europe requires, and what it never required
The obligation to check energy certificates is European, not national. It begins with Directive 2010/31/EU on the energy performance of buildings, whose Article 18 and Annex II create an "independent control system": each member state must verify a sample of certificates, at one of three possible levels — validation of input data, full verification of the calculation, or a visit to the building.
One detail shaped twenty years of practice: the 2010 directive asked for a "statistically significant" sample, and Directive (EU) 2018/844 deleted the word "percentage" from the text. Yet "statistically significant" has a precise meaning. With the parameters normally used across European practice, relayed by the Concerted Action EPBD — a 5% margin of error, a 95% confidence level — a country issuing one million certificates a year satisfies the obligation with about 384 checks, or 0.04%.
In other words: the British 2% was never imposed by Europe. It is a national choice, more demanding than EU law. And France's 0.08% in 2023 was, against the obligation as it stood until 2026, formally sufficient. Criticising France for failing to meet a 2% rule means criticising it for not applying a rule that did not exist.
This is precisely what the 2024 recast corrects. Directive (EU) 2024/1275, at Article 27 and Annex VI, replaces the obligation of means with a quantified obligation of result. Each member state must now guarantee that at least 90% of the valid certificates issued are compliant, with a 95% statistical confidence level, over a period not exceeding one year. Verification must rest on random sampling, of which at least 10% must involve an on-site visit — new in the recast. Where the system is delegated to non-governmental bodies, at least 25% of the sample must be verified by a third party. And the results must be published.
The date to remember: 29 May 2026. That is the transposition deadline for Directive (EU) 2024/1275 (Article 35). It has passed. On 15 July 2026 the European Commission sent letters of formal notice to all 27 member states, without exception, for incomplete transposition. So this is not a French delay: it is a generalised European one. Note that several professional analyses have circulated a "May 2029" deadline: it is wrong.
One sentence in Annex VI deserves particular attention, because it speaks directly to the route France has chosen: targeted analysis of the data "cannot be used as a basis for measuring the overall quality of the system". Detecting suspect certificates and measuring the quality of a stock are two different exercises. The directive now requires both.
Three architectures of control
The British model: auditing certificates
In England and Wales, an assessor may only lodge a certificate as a member of an accreditation scheme — a private body approved by the housing ministry. There are six of them today. Their approval comes with written conditions, the Scheme Operating Requirements, which set the audit regime. The ministry's own document is one line long: schemes must audit at least 2% of the certificates lodged in each strand, of which "a significant proportion" must be randomly sampled — no percentage is put on it. The industry document that implements it adds two per-member floors: every member must have at least 0.5% of their lodgements randomly audited each year and, for assessors working on the existing stock, at least one audit every twelve months. An audit fails as soon as the sum of the discrepancies found field by field exceeds 5 SAP points; the certificate is then declared defective, must be replaced, and a follow-on audit is triggered.
One point is regularly misreported: this selection is not purely random. Since 2019 — after a consultation in which 85% of respondents backed the change — it combines a random base, follow-up audits after a failure, and risk-targeted audits, triggered by around thirty rules, of which some twenty are live. Funding is direct: schemes levy a fee on every certificate lodged, in the order of £3 to £7.
For roughly 1.75 million residential lodgements a year, the 2% floor represents close to 35,000 audits annually. That is the order of magnitude to keep in mind for what follows.
The French model: supervising professionals
France made a different choice, and an old one. Article L. 271-6 of the Construction and Housing Code requires the assessor to be certified by an accredited body, insured and independent. The state does not accredit assessors: through Cofrac, the national accreditation body, it accredits thirteen private certification bodies, which in turn certify individuals.
The order of 20 July 2023 sets the tempo: a seven-year certification cycle comprising six surveillance operations — three documentary reviews (in years 2, 4 and 6, on a sample of at least five reports) and three on-site checks (year 1 as supervised practice, years 3 and 5 as counter-expertise). Contrary to a common belief, half of these operations therefore involve a site visit.
In 2023 this system produced 2,882 documentary reviews and 937 on-site checks — 3,819 checks in total — with 18% of them finding breaches, 1,049 certification suspensions (11% of all certifications) and 692 withdrawals (7%). Set by the Court against the annual volume of residential certificates, that gives the well-known 0.08%. But set against the population of assessors — the object actually being audited — the sanction rate is anything but marginal: it is the backdrop to the crackdown on the profession that has followed.
And France has since changed gear. Two orders have built a third way:
- The order of 16 June 2025 introduced, from 1 October 2025, statistical screening of the entire ADEME database: analysis of assessors' activity triggers an automatic referral to the certification bodies, which must open a check within one month.
- The order of 28 July 2025 codified the first anomaly indicator: crossing 1,000 house or flat certificates over a rolling twelve months, a threshold beyond which the workload is deemed physically impossible.
This pairing — exhaustive screening plus a duty to act within a month — has few equivalents in Europe. It is the main strength of the current French system, and it is rarely mentioned. Its limitation is symmetrical: only one indicator is codified. The other criteria belong to an unpublished operational doctrine, which rules out any democratic scrutiny of what is actually being looked for — and, as we shall see, does not meet the European requirement to measure.
What everyone else does
Eight European systems, compared on what matters: who audits, what, how much, how, and with what consequences.
| Country | Object audited | Published coverage | Selection | Published error rate |
|---|---|---|---|---|
| France | The professional (6 operations / 7 years) + screening of 100% of the database since 01/10/2025 | 0.08% of certificates (2023) | Mandatory cycle + automatic referral | Not published |
| England & Wales | The certificate | ≥ 2% of lodgements / 12 months | Random + risk targeting since 2019 | Not published by the schemes |
| Denmark | Automatic validation of 100% (340 checkpoints) + thematic audits | ~0.24% in in-depth audit | Targeted by annual risk theme | Yes, long series (31% in 2016, random) |
| Netherlands | The certified company + case files | 2% up to 1,000 labels, then 1 in 1,000 | Sample per certifier | Yes: 8.3% critical deviation in housing (2024) |
| Flanders | The certificate, with on-site re-measurement | 2,207 units and 459 experts audited (2024) | Random + targeting + anomaly detection | Partial (targeted sample) |
| Ireland | The assessor, with a guaranteed floor | ≥ 2 audits per assessor per year | Random + targeting | Not published |
| Italy (Lombardy) | Automatic checking of 100% + documentary sample and visits | ≥ 2% (national obligation) | Automatic + stratified sample | Not officially published |
| Germany | The certificate | Set by each Land, not published | Purely random draw | Not published |
Three lessons emerge from this table, and none of them points where you would expect.
First lesson: the lever that produces results is targeting, not volume. Flanders went from 3 expert suspensions in 2023 to 26 in 2024 without any dramatic increase in audit volumes, simply by introducing anomaly detection into its selection. That is exactly the bet France made in October 2025.
Second lesson: the cheapest check is the one embedded in the software. Denmark automatically validates 100% of lodged certificates against 340 checkpoints, before any question of human audit. Lombardy does the same. A consistency check run at data entry costs a fraction of an audit — and prevents the error instead of recording it.
Third lesson, the most uncomfortable: almost nobody publishes. Ireland probably has the best-designed system in Europe — at least two audits per assessor per year guaranteed in writing, a two-year points system, automatic revocation of the certificate for serious breaches — and publishes no results at all. Germany does not even publish the percentage audited. Only Denmark and the Netherlands provide usable series. The finding that holds for the whole of Europe is therefore this: almost no country can say what percentage of its certificates is correct. That is precisely what the 2024 directive intends to end.
The same home, five assessors
First signal: dispersion between assessors
There is one test every country has ended up running, and it shifts the debate: send several professionals into the same home and compare the results.
In France, the consumer body UFC-Que Choisir commissioned 34 certificates across 7 houses in September 2022: six houses out of seven did not receive the same class depending on the assessor, with up to three classes of difference and consumption estimates varying threefold. The same year, the magazine 60 Millions de consommateurs found, across four homes visited by five professionals, "always at least two different letters, sometimes three".
The result is in no way French. In Belgium, an investigation by public broadcaster RTBF had the same flat certified by five certifiers: from 179 to 264 kWh, a 47% spread — and consumer group Test-Achats noted that the same 47% spread had already been measured ten years earlier. In the Netherlands, a study commissioned by the ministry in 2022 produced eleven labels across four homes: three homes out of four received different letters. In the United Kingdom, a mystery-shopper exercise covering 29 homes, each assessed four times, found an average gap of 11.1 points between the highest and lowest rating for the same property, with close to two-thirds of homes varying by at least two bands (Jenkins, Simpson and Peacock, as reported by Crawley et al., Energies, 2019).
These four countries have very different control systems and obtain the same dispersion. That is the single most important fact in this article. The main problem is not the audit rate: it is the variance of field data entry — measured areas, wall type, floor type, insulation thicknesses, default assumptions. British research identifies exactly the same causes. An audit records the error; it does not remove what produces it.
Second signal: bunching at the thresholds
Dispersion is not the only observable signature. There is a second, more specific one: the way labels pile up just on the right side of a regulatory boundary. A certificate that moves from F to E lifts a home out of the rental ban; the incentive is direct, and it leaves a statistical trace.
In France, this effect is measured. The Council of Economic Analysis estimated in June 2024 that 3.9% of certificates were "suspected of being manipulated at the thresholds" before the 2021 reform, and 1.7% over the period from July 2021 to December 2023 — a reduction of more than half, across the D/E, E/F and F/G boundaries. Set against annual volume, that still represents on the order of 68,000 certificates a year.
The same phenomenon shows up in the United Kingdom, where the structure of the score — a 1-to-100 rating cut into bands — makes the step particularly visible just below a band boundary; our British platform has documented it on the open English database. On the French side, by contrast, no equivalent population study has been published: bunching is measured by the Council on an aggregate indicator, never mapped boundary by boundary as it is across the Channel. That is a documentary blind spot, not a methodological one.
Two distinct problems are constantly conflated. Bunching at the thresholds is measured, and it has receded: 1.7% of certificates since July 2021, against 3.9% before (Council of Economic Analysis, 2024). Dispersion between assessors is of an entirely different order of magnitude: up to three classes apart on the same home. The first is a matter of auditing and sanction; the second is a matter of method, training and field data entry. Strengthening audits acts on the first, much less on the second.
Five myths about auditing energy certificates
"Europe requires X% of certificates to be audited"
No, and not since 2018. The word "percentage" disappeared from the text, replaced by a statistical requirement that a country issuing one million certificates met with fewer than 400 checks. It is the 2024 recast that reintroduces figures — but in the form of a validity rate to reach, not an audit rate to perform.
"There are no on-site checks in France"
False: three of the six operations in the certification cycle involve an on-site check. By way of comparison, the German regime makes visiting the building conditional on the owner's agreement.
"A high error rate proves the system is bad"
Usually the opposite. Denmark found roughly 31% of labels in error under random sampling in 2016; after switching to risk-targeted selection, its detection rates exceeded 90%. The Danish Energy Agency says so itself: the figure "does not indicate the general quality of Danish labels, but rather the effectiveness of the control". A rate drawn from a targeted sample measures the quality of the targeting, not the quality of the stock — and that is exactly why the 2024 directive forbids using it to assess a system.
"Wrong certificates cost billions in public subsidies"
A fragile shortcut. For whole-house renovations, grant assessment relies mainly on the energy audit, not on the certificate. And the figures regularly quoted correspond to sums detected and prevented before payment, not to losses actually incurred. The reliability of the certificate is a real issue; it does not need that number.
"France is Europe's poor performer"
The compliance study carried out for the European Commission in 2015 in fact ranked the French system among the most robust, alongside Denmark, Italy, Wallonia, Cyprus and Lithuania. France sits in the European average on auditing, above it on detection since October 2025, and below it on measurement and transparency. It is that imbalance that is the problem, not a general backwardness.
What stronger auditing would cost
The question is less open than it looks, because the administration has already answered it. In the impact assessment for the order of 20 July 2023, the French housing directorate put the cost of strengthening the certification system at "an extra cost of around €5 per certificate for initial certification and €3 per certificate for renewal, that is an extra cost of 2% to 3%".
A second anchor comes from Scotland, the only territory to publish the cost of its audit function: £428,000 a year for audit and inspection, covering roughly 205,000 certificates, or £2.09 per certificate — funded by a lodgement fee raised from £2 to £6, which the Scottish Government describes as "less than 10% of the total cost of an EPC".
From there, an order of magnitude can be built for France. Our assumptions, stated explicitly: four million certificates a year, an average price of €175 (the midpoint of the €150-250 range), a 2% target (the British standard), and a unit audit cost between €120 (industrialised documentary check) and €500 (on-site check). Result: 80,000 audits a year, for an extra cost of between €2.40 and €10 per certificate, or 1.4% to 5.7% of its price. The central scenario lands around €6 per certificate, roughly €24m a year.
Three of the four assumptions above are ours. This is not a measurement, it is an order of magnitude — bracketed, however, by three independent sources: the housing directorate (2% to 3%), Scotland (£2.09, our low scenario) and Ireland (a €30 per-certificate fee funding the entire public system). The calculation ignores the cost of building the system, the deterrent effect — probably the main channel of effectiveness — and supply capacity: 80,000 audits a year presuppose an audit workforce that does not exist today.
The conclusion of this section is therefore not budgetary. The blocker is not cost: every public source converges on 1% to 6% of the price of a certificate. The blocker is architectural. Supervising professionals at a rate of six operations every seven years cannot, mechanically, produce high coverage per certificate, whatever the budget. And France is the only one of the systems compared here without a per-certificate fee: its auditing is funded by a fixed cost per professional, identical for someone producing 100 certificates a year and someone producing 2,000.
One further element must be added, one the Court of Auditors states plainly and that an article on certificate quality cannot pass over: the average price of a certificate "does not always allow the assessor to be paid in line with the time that would be needed to obtain a quality result". The same Court judges proven fraud to be "marginal" and attributes the breaches observed above all to newcomers' unfamiliarity with the rules. Strengthening auditing without addressing the economics of the service would mean penalising the effects of a constraint left untouched.
Check your certificate for inconsistencies
As long as public auditing covers 0.08% of certificates, the first check on a certificate remains the one its recipient carries out. Our tool applies a series of consistency rules to the public data of your certificate — label against declared consumption, heat loss against floor area, thermal compliance against year of construction — and flags the gaps. It is the method we applied at scale in our reliability barometer. It is a warning signal, not a verdict: it replaces neither a regulatory check nor the assessor's professional opinion.
What is going to change
Three things, and they are coming fast. First, the European obligation changed in kind on 29 May 2026: states are no longer asked to audit a sample, they are asked to reach and publish a validity rate. Second, France finds itself in a singular position: it has built detection — exhaustive screening of the ADEME database, with a duty to act within a month, puts it ahead of most of its neighbours — but not measurement, that is, the properly sized random sample the directive now makes the condition of proof. Both are required, and Annex VI is explicit: targeting cannot stand in for measurement.
Finally, 2026-2027 will be the first period in which comparable validity rates across countries become public. That will be the real novelty: not knowing who audits the most, but knowing, for the first time, what percentage of certificates is correct. Until that publication, verification remains largely the responsibility of whoever receives the document — which is exactly what our verification tool is for.
