FR Newsletter

AI for Work

Replying to customer reviews with AI, without crossing into fake reviews

There are two very different things people mean by "using AI for reviews". Replying to the reviews you received, faster and more calmly: useful, allowed, and a good use of the tool. Producing the reviews themselves, buying them, or quietly burying the bad ones: a criminal offence with a named penalty. The line is sharp, so here it is, along with a method for the legal half.

The four things the law forbids

The DGCCRF, France's consumer protection authority, sets out the rules in its practical sheet on online reviews. A professional giving access to consumer reviews must state whether and how it guarantees that published reviews come from consumers who actually used or bought the product. And the consumer code prohibits a professional from:

  • falsely presenting themselves as a consumer;
  • claiming that reviews come from consumers who used or bought the product without having taken the necessary steps to verify it;
  • publishing, or having someone else publish, fake reviews;
  • modifying consumer reviews in order to promote products.

The penalty is not symbolic. These acts constitute the offence of misleading commercial practices, punishable by two years' imprisonment and a 300 000 euro fine, an amount that can be raised, proportionately to the advantage gained, to 10 percent of average annual turnover or 50 percent of the expenditure incurred for the practice. Knowingly providing help or assistance carries the same penalties.

That last sentence is the one to read twice if you were planning to outsource anything. The DGCCRF's investigations found fake reviews written by the professional, by people around them, or by third parties, including specialised providers with a know-how in writing style and false identities. It also reports a newer practice: recruiting ordinary consumers on social networks or messaging apps to post a positive review in exchange for a free product or other compensation. The DGCCRF is explicit that the professional, the provider and the recruited individuals all incur criminal liability.

Screenshot coming after our hands-on test
A reply drafted, corrected and posted: the three versions. Screenshots coming with our hands-on tests.

Deleting the bad ones counts too

The temptation is rarely "let's buy reviews". It is usually "can we get this one taken down". The DGCCRF's investigations list biased moderation as a practice it has found: deleting all or some negative reviews, or delaying the publication of negative reviews, in order to skew the ranking.

Reporting a review that genuinely breaks a platform's rules is a legitimate action. Filtering your own display of reviews so the negative ones disappear is a different act, and it is on the list.

The platforms add their own rules

Google's content policy for Maps starts from one principle: contributions must reflect a genuinely lived experience at a place or business, and reviews and ratings must be authentic. Artificial engagement is not allowed and will be removed. The examples it gives are worth knowing by heart:

  • content not based on a real experience, or that does not faithfully represent the place or product;
  • reviews or ratings paid for, directly or in kind;
  • content posted from several accounts by the same person, or at their request;
  • content posted using an emulator or another circumvention service.

Asking a real customer to leave a review is normal. Offering them anything for it is not, and it is the single most common mistake made in good faith.

What the AI Act does and does not require here

This is worth being precise about, because the internet is full of confident claims in both directions. Article 50 of the AI Act requires disclosure in specific situations: when people interact directly with an AI system, for deepfakes, and for AI-generated text published to inform the public on matters of public interest, unless it was human-reviewed under someone's editorial responsibility.

A reply to a customer review, signed by your business, is none of those. So the AI Act does not, on its own, oblige you to label a review reply as AI-assisted. What does apply, in full, is consumer law: the reply must be true, must not pretend to be a consumer, and must not misrepresent what happened. If you automate replies through a conversational agent that customers interact with directly, you are in a different situation, and Article 50's first paragraph on informing people they are dealing with an AI system comes back into play.

The method for the legal half

The same rules on honest commercial writing run through our guide on writing a product page with AI. What an assistant is genuinely good at here is tone under pressure, consistency across fifty replies, and not writing the sentence you would regret.

Here is a customer review of my business, and the facts I know about what happened. Draft a public reply of 60 words maximum. Use only the facts I give you. Do not invent a cause, a gesture, a discount or an apology for something I have not confirmed. Do not promise anything. Thank them, acknowledge the specific point they raise, say what we are doing about it if I told you, and invite them to continue by [channel]. No marketing language.

Four rules that make the output usable:

  1. Facts in, facts out. The assistant does not know what happened in your shop. Everything factual must come from you.
  2. No promises. A refund or a gesture invented by a model becomes a commitment in public.
  3. Short. Sixty words read as sincere; three hundred read as legal.
  4. One human reads it before it goes out. Always, including on the fiftieth.

The second genuinely valuable use is the one nobody talks about: paste your furious draft and ask for the same content with no emotion, no accusation and no sarcasm. The reply you do not send is worth more than the one you do.

The negative review you deserve

If the review is accurate, say so. Acknowledging a real failure in public, briefly and without excuses, is the most persuasive thing a business can write, and no tool can fake it. Where an assistant helps is in stopping you from doing the three things that make it worse: arguing the facts in public, questioning whether the person is a real customer, and explaining your internal organisation at length.

The line, in one line

Write replies: yes, with facts you own and a human check. Write reviews, buy them, trade them for a discount, or prune the bad ones: that is a misleading commercial practice with a two-year prison sentence and a 300 000 euro fine attached, and the penalties reach whoever helped. We have run no hands-on test on review management tools, so this guide names none and recommends none.

Tools in this guide

Questions

Is it legal to write review replies with AI?

Replying to a review you actually received, in your own name as the business, is ordinary customer correspondence. What the French consumer code prohibits is presenting yourself falsely as a consumer, publishing or having published fake reviews, claiming reviews come from verified customers without taking the necessary steps, and modifying consumer reviews to promote products.

What are the penalties for fake reviews in France?

The DGCCRF states that these acts constitute the offence of misleading commercial practices, punishable by two years' imprisonment and a 300 000 euro fine, which can be raised proportionately to 10 percent of average annual turnover or 50 percent of the expenditure incurred for the practice. Knowingly providing help or assistance carries the same penalties.

Can I offer a discount in exchange for a review?

Not on Google. Its Maps content policy does not allow merchants or users to solicit or encourage content that does not represent a genuine experience, nor to offer incentives such as payment, discounts or free goods and services in exchange for posting a review, revising one, or removing a negative one.

Sources

  1. DGCCRF, Avis en ligne : attention aux faux commentaires !
  2. Google, Maps user-generated content policy, prohibited and restricted content
  3. Regulation (EU) 2024/1689 (AI Act), Article 50, transparency obligations, EUR-Lex