Energy: How AI Recommends an Electricity Supplier

Ask ChatGPT which electricity supplier to choose, and it will almost never give you a single name: it answers by category (price, customer service, green offers), names several providers, and avoids a definitive ranking. This is not editorial caution: it is that the French market changes at least twice a year by regulatory decision, and commercial offers can shift monthly. For an energy supplier, being visible in these answers depends less on brand recognition than on how fresh and verifiable what it publishes actually is.

This article covers what AI engines actually answer about electricity suppliers, why the energy ombudsman's official comparator is a source AI should cite more often, why tariff freshness is a structural problem for a language model, and what a supplier or comparator can do to be cited correctly. The French market is the focus throughout, since the rules and comparators described here are French.

What does an AI actually answer when asked which electricity supplier to pick?

It answers, but in a segmented and cautious way, naming several suppliers rather than a single "best" one. Selectra asked ChatGPT-5 on September 15, 2025, about "the best electricity supplier": the model answered by criterion (price, customer service, green electricity), named several suppliers depending on the category, and advised against others, without giving a clear-cut verdict (Selectra, 2026). Selectra, itself a commercial comparison site, also notes that an AI's reliability on this kind of question should still be treated with caution, even when the model has web access.

The answer has evolved over time. According to the same source, in 2023 ChatGPT mostly named the historical incumbents (EDF, Engie, TotalEnergies, Eni) when asked a comparison question. Three years later, its answers appear more segmented, a sign that the model is drawing on more recent third-party content rather than a handful of well-known brands. This is the same mechanism we describe in how ChatGPT chooses its sources: the more up-to-date and specific content exists on a topic, the less an answer sticks to the historical incumbents.

Another notable shift: commercial comparators no longer settle for being cited, they are embedding themselves directly into the interface. LeLynx.fr announced, in a press release dated May 11, 2026, a ChatGPT app that generates an electricity quote without leaving the conversation: the user just needs to provide the meter's power rating, annual consumption, and desired offer type (LeLynx, 2026). For a supplier, this means part of the comparison no longer happens on a traditional website, but inside an AI assistant's chat history.

The energy ombudsman's official comparator: a source AI should rely on more

There is a public, free, independent comparator backed by a legal mandate, and most AI engines asked about electricity do not mention it on their own. The comparator run by France's national energy ombudsman (comparateur.energie-info.fr) is the only public comparator for energy offers, anchored in the French energy code since the 2019 energy-climate law. It lists offers from roughly forty national suppliers and about a hundred local distribution companies, for individuals and small businesses (energie-info.fr).

What sets it apart: it does not just display a price, it also shows a one-to-four-star service quality indicator, calculated from the rate of complaints that led to an ombudsman referral for each supplier. The ombudsman itself is explicit that it never recommends one supplier over another: the tool compares, it does not advise.

Commercial comparators (Selectra, LeLynx, Hello Watt and others) answer to a different framework: the French consumer code (articles L. 111-7 and D. 111-11) requires them to disclose their ranking criteria, their financial relationships with the suppliers they list, and how that listing works. The French consumer protection authority (DGCCRF) relies in part on referrals from the ombudsman and from the energy regulator (CRE) to target its inspections, and has already fined one energy comparator €400,000 in 2024 for misleading practices in its subscription journey (French Senate, ministerial reply, 2025). A supplier working on its AI visibility therefore benefits from being present on both fronts: the official comparator, which carries no commercial bias, and private comparators, whose own transparency about how they rank offers is itself a trust question, much like what we describe for another heavily-compared sector in banking and insurance facing AI.

Why electricity prices are a freshness problem for AI

Electricity prices change by official decision at least twice a year, which means an AI-generated answer can be out of date the moment it is written. The regulated wholesale access to historical nuclear power (ARENH), which since 2011 had capped the price alternative suppliers paid for electricity, ended for good on December 31, 2025. Since then, the regulated electricity tariff (TRVE) has been calculated from wholesale market prices instead.

A direct, dated consequence: the French energy regulator (CRE), in its deliberation n° 2026-06 of January 14, 2026, proposed keeping the TRVE broadly stable as of February 1, 2026, with an average decrease of 0.83% including tax, as supply costs fell 4.22% before tax compared with 2025 because of this change in calculation (CRE, deliberation 2026-06; press release of January 20, 2026). The same regulator then proposed an average 2.5% increase for August 1, 2026. Two official decisions pointing in opposite directions within six months, on a tariff that only covers the regulated share of the market — market offers can move even more often.

A language model trained on data frozen at a given date cannot track this calendar. It can cite a price that was accurate the moment it was trained and wrong six months later, or credit a scheme (ARENH) that no longer exists. It is the same freshness principle we cover in why AI engines love dated figures: a number without a reference date loses its evidentiary value, and that is especially true in a sector where the regulator changes the calculation rule several times a year. Our calibration study from September 30, 2026, actually treats freshness as a standalone criterion in the AI-visibility score (methodology and calibration study), precisely because pages with an explicit update date are the ones AI engines can date, and therefore use with more confidence.

The national energy ombudsman's 2025 annual report, published in 2026, illustrates this price effect on disputes: 10,475 referrals received in 2025, down roughly 10% from 2024, of which 6,545 were deemed admissible. The ombudsman attributes this drop to falling wholesale prices and a 15% cut to the regulated tariff in February 2025, which brought price-related complaints down 62%. The remaining disputes are now more technical (grid connection, solar, bottled LPG), and the amicable settlement rate fell to 49% from 58% the year before (energie-info.fr, 2025 report). The ombudsman also hands out "red cards" to the most problematic suppliers: in 2025, Primagaz (propane gas) and JPME, an electricity supplier whose authorization to buy for resale was withdrawn in January 2026 (UFC-Que Choisir, 2026). An AI that does not read this annual report has no way of knowing a supplier lost its authorization to operate — one more reason for a serious supplier to document its own service indicators, with a date, rather than leave the gap to be filled by outdated training data.

Green electricity: what AI can (and cannot) verify

A "green electricity" offer is a claim no AI can verify physically, which makes it fertile ground for vague marketing when a supplier does not publish clear proof. On the power grid, it is physically impossible to tell whether the electron delivered to a given customer came from a wind turbine or a gas plant: every electron mixes together. A supplier can legally call itself "100% green" by buying guarantees of origin, certificates governed by EU law that certify an equivalent amount of renewable electricity was fed into the grid somewhere else, with no direct link to what the customer actually consumes.

To reduce this ambiguity, the French environment and energy agency (ADEME) created the VertVolt label, launched in late 2021. It has two tiers: the "committed" tier requires the supplier to buy, from French producers, an amount of renewable electricity equal to what it sells to its labeled customers, guarantees of origin included; the "highly committed" tier adds that at least 25% of that electricity must come from recent installations that receive no public subsidy or are community-governed, and excludes nuclear power from the labeled scope. The label also requires the supplier to disclose the production technology and geographic origin, as well as the share of its customer base on the labeled offer (ADEME, VertVolt).

For an AI just as for a buyer, what separates a verifiable claim from a marketing line is the presence of dated, sourced proof rather than a phrase like "100% green electricity" with no reference. It is the same principle we describe for expertise claims in E-E-A-T and AI visibility: a named, dated third-party certification carries more weight in a generated answer than a self-declared one.

What should you do in practice?

For an energy supplier or comparator, visibility in AI answers is built less on brand recognition than on how verifiable and how fresh what is published actually is.

1. Date your tariff pages explicitly. A price grid with no update date cannot be used with confidence by an AI, especially in a sector where the regulator changes its calculation rule at least twice a year. State the date of the last revision and, where possible, reference the matching CRE deliberation.
2. Cite the official comparator instead of ignoring it. A supplier that links to comparateur.energie-info.fr on its own pages gains credibility: it shows it has nothing to hide from a source with no commercial bias.
3. Publish your service indicators, not just your prices. Complaint rate, response time, support availability: dated, verifiable figures that echo what the ombudsman publishes every year for the whole market.
4. Back up a green offer with third-party proof. VertVolt tier, the real share of French guarantees of origin, production technology: these are verifiable elements, unlike an advertising slogan.
5. Disclose your ranking criteria if you run a comparator. It is a legal requirement under the French consumer code, and it is also what lets an AI understand why one supplier ranks above another.
6. Measure regularly what AI engines say about your brand against other suppliers, using the same method each time: our guide to measuring AI visibility compares the available approaches.

For a full measurement that compares your supplier or your offer with direct competitors on the questions consumers actually ask (price, green electricity, switching supplier), Beeleven's AI visibility audit asks ChatGPT, Gemini, Claude and Perplexity these exact questions and measures who is cited, in which position, and against which competitors. It is the same third-party-proof logic we cover in sources and AI citations: what a third party verifies and dates carries more weight than a claim made on your own site.

Frequently asked questions

Can an AI really tell you the best electricity supplier?

It can name several suppliers by category (price, service, green offers), but it does not settle the question like an official ranking, and its answer depends on how fresh the content it could access actually is. According to Selectra's September 2025 test, ChatGPT-5 answers in a segmented way rather than with a single name.

Is the energy ombudsman's comparator more reliable than a commercial one?

They serve different roles. The public comparator has no commercial relationship with suppliers and makes no recommendation; private comparators are paid by the suppliers they list, which is why they face transparency requirements on their ranking criteria (French consumer code) and oversight from the DGCCRF.

Why can ChatGPT give an outdated electricity price?

Because the regulated tariff is revised by the CRE at least twice a year, and market offers change even more often. A model trained on data frozen at a given date cannot keep up with that pace without access to dated, up-to-date sources.

Is a "green electricity" offer cited by an AI necessarily trustworthy?

No. A "100% green" claim rests on guarantees of origin, which do not mean the electricity actually delivered physically came from a renewable source. ADEME's VertVolt label, with its two tiers, is the most verifiable reference for telling a genuinely committed offer apart from a marketing line.

Sources

• French national energy ombudsman, on the relaunch of its offer comparator, energie-info.fr (in French).
• French national energy ombudsman, 2025 annual report, energie-info.fr (in French).
• CRE, deliberation n° 2026-06 of January 14, 2026 on proposed regulated electricity tariffs (in French).
• CRE, press release of January 20, 2026 (in French).
• ADEME, VertVolt label (in French).
• French Senate, ministerial reply to Senator Maurey's question, 2025, on DGCCRF oversight of energy comparators (in French).
• LeLynx.fr, press release on its ChatGPT app for electricity quotes, May 11, 2026 (in French).
• Selectra, test of ChatGPT-5 on the best electricity supplier, 2026 (in French).
• UFC-Que Choisir, "Red cards despite an overall drop in disputes in 2025", 2026 (in French).

→Banking and Insurance: What ChatGPT Recommends to the French

→How AI Engines Choose the Products They Recommend