Should you tell customers a reply was written by AI

No platform we read requires you to tell a customer that a reply was written by AI. The nearest thing to a rule is on Instagram, where Meta says an automated chat experience must disclose itself, but only “When required by applicable law”. On WhatsApp the same company asks for nothing of the kind: we searched its policy and its terms and found no disclosure requirement. Google’s review content policy pulls AI-generated content into scope and then asks for no label anywhere. Two Meta-owned platforms differing is the most useful thing here.

Instagram: Meta states the obligation and defers on whether it applies

Meta’s overview of the Instagram Messaging API, read by us on 19 September 2026, carries the clause:

When required by applicable law, automated chat experiences must disclose that a person is interacting with an automated service: at the beginning of any conversation or message thread, after a significant lapse of time, or when a chat moves from human interaction to automated experience.

Read the first five words again. Meta does not mandate disclosure; it mandates disclosure where the law does, then names the three moments at which it has to happen. It also points at where to look, listing groups that “should pay special attention to this requirement”: the “California market or California users” and the “German market or German users”. Sample wording is offered too, such as “I am an automated chatbot.”

Meta then goes past its own rule:

Even where not legally required, we recommend informing users when they’re interacting with an automated chat as best practice, as this helps manage user expectations about their interaction with your messaging experience.

A consumer help page puts the rule to users in plainer words, “Where legally required, these automated chats must tell you when you’re interacting with them”, and covers the voluntary case: “Instagram accounts can also remind you when you’re messaging with an automated chat even when it isn’t legally required.”

WhatsApp: no disclosure requirement that we can find

We expected symmetry here and did not find it. We read WhatsApp’s Business Messaging Policy and the WhatsApp Business Solution Terms the same day, searching both for the obligation above. The words disclose, automated and artificial intelligence do not appear in the policy at all, and automation appears once. In the terms, artificial intelligence appears four times, all inside the AI Providers clause about vendors and training data, and the one use of disclosure is a data-security sentence about unauthorised processing. Neither document asks a business to identify an automated service to the person it is messaging.

What the policy does require is the exit:

You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths.

It enumerates them, beginning with “In-Chat Human Agent transfer” and including a phone number, email, web support and an in-store visit. Instagram states it more bluntly: “Apps that use the Instagram Messaging APIs must have an escalation path to a human agent”, and an experience may start from automation but “must have a way for users to chat with a human agent as needed”.

So on both channels the route to a person is unconditional and the label is not. Meta treats being stuck with a machine as the harm, not being answered by one. On the channel this product uses, whether you say a reply was automated is local law’s question, not Meta’s.

Google puts AI content in scope and never asks you to label it

The scope clause is one sentence, on Google’s About our policies page:

These policies and standards apply to all Maps user generated content, regardless of whether the content is human-generated or AI-generated content, including:

That sentence is the whole of Google’s engagement with the question in the pages we read. Across five Google help pages, three in that policy section and two in Business Profile Help, the word AI appears once, in that sentence. The word disclose appears zero times, and the one instance of disclosure sits in the definition of personal information.

Your reply is moderated against that policy: “Google reviews your replies: This makes sure that they follow Google’s content policies.” A generated reply is judged by the same rules as a typed one, with no separate standard and no labelling duty attached, and a rejection arrives through the published rejection reasons.

Google ships its own drafting tool, and we found nothing about marking its output on the help page documenting it. That page lists, among what the integration can do, “Engage with customers by drafting replies to reviews and creating posts”. The absence is a finding about the five Google pages we read, not a permission: on those pages Google has not decided this for you, and a vendor claiming Google requires an AI label should be asked which page says so.

Google does label its own generated text

The one place we found Google publishing an AI disclosure is where Google is the author. Its Places API documents a disclosureText field on those summaries:

The AI disclosure message “Summarized with Gemini” (and its localized variants).

The same field hangs off the object Google calls an “AI-generated summary of the place using user reviews”. So Google labels text Google produced about a business and requires no label on text the business produces about itself. That is API documentation rather than merchant guidance, and we read it as a contrast, nothing more.

Telegram discloses structurally instead: its developer introduction says “Bot usernames normally require a ‘bot’ suffix”, then names accounts carrying none.

We are not going to summarise the law, and you should check yours

Meta’s clause defers to applicable law, so the binding rule for you is not on any platform’s page. We are not going to summarise it here: not for California, not for Germany, not for the EU, not for anywhere else. We are not lawyers, the rules move faster than this page will, and a wrong summary of a disclosure duty is worse than none because it reads like permission. Meta names those two markets itself. Use that as a pointer to ask someone qualified where you trade.

Our position, which is ours and not a rule

Everything above is what the platforms say. This paragraph is ours. The reply publishes under the business’s name, and the owner is accountable for it whichever tool drafted it. Google describes the byline plainly: “It will appear like your business replied, and your personal name won’t be shown.” A reply you did not read is still a statement you made, so our position is that approval, not a label, is what makes a drafted reply honest. In a live chat, unlike a published reply, tell people anyway: Meta recommends it where no law compels it, and the real complaint is being unable to reach a person.

mesej is specified to draft replies for an owner to approve rather than to post them, and nothing of ours is connected to a channel or a Business Profile today, so no reply of ours has published anywhere and we have no measurement of how customers react either way. What to write is covered in answering a negative review, and the tool category in review management software.

The short version

On Instagram, Meta requires an automated chat to disclose itself only where law requires it, names three moments, and recommends it anyway. On WhatsApp we found no disclosure requirement, only the human escalation path both channels require unconditionally. Google pulls AI-generated content into its policy scope, moderates your reply against it, drafts replies through Gemini and asks for no label at any point, while shipping one for its own summaries. So on the platform pages listed below, and on those pages only, we found nothing requiring you to mark a review reply as AI-written. That is a finding about what those platforms publish. It is not a finding about the law, which this page does not summarise for any jurisdiction and will not: Meta’s own clause defers to it, so whether anything obliges you is a question for somebody qualified where you trade. On the platform question, decide it on what your customers would want to know.