Summary

  • ComReg 26/66 recognises that AI can help consumers, business users and providers organise or clarify dispute submissions, including where language or accessibility is a barrier.
  • The notice leaves accuracy with the submitting party. It identifies fabricated or incomplete facts and references, inaccurate legal citations and excessive length as risks, and says ComReg may seek clarification or a shorter or revised submission.

A case file can be easy to read and still be difficult to verify. A page of fluent prose may describe a billing error, a missed repair or a damaged parcel in a way that sounds complete. The decision-maker still needs to know which date came from the provider’s message, which amount came from the bill, and which conclusion is the submitter’s interpretation. ComReg’s new notice on artificial intelligence is about that gap between a persuasive account and a checkable record.

ComReg lists its “Guidance on the use of AI in end-user dispute resolution” as Information Notice 26/66, posted on 6 October 2026. Its news announcement, dated 8 October, says the guidance covers consumers and business users of phone, broadband and postal services, as well as providers that are parties to disputes. The regulator acknowledges that AI can help prepare submissions, organise information, improve clarity and address language or accessibility barriers. It also warns that generated material can be inaccurate, misleading, incomplete or fabricated.

That combination matters. A categorical ban could remove a tool that helps someone turn scattered emails, invoices and service records into an intelligible chronology. Uncritical acceptance would let a confident-sounding system supply facts or legal citations that the person cannot substantiate. ComReg’s public summary instead places responsibility on the party using the tool: check the facts, verify references, and ensure the submission is accurate and relevant. It also says AI use in submissions should be transparent; the public summary does not specify a standard disclosure form.

In appropriate cases, the regulator may ask how AI was used or request a different, shorter or revised format.

The dispute file already has a sequence

For electronic communications, the AI notice sits inside a process that already separates an initial complaint from formal resolution. ComReg’s public information describes a free Consumer Care service and a more formal end-user dispute procedure. An end-user must first use the provider’s complaint process; the statutory route allows referral after at least ten days from the complaint or once the provider’s code-of-practice procedure is complete. The accepted formal case is no longer handled through the informal Consumer Care channel.

The 2023 Act gives ComReg, or an independent person it appoints, the role of proposing a resolution. The end-user decides whether to accept it. If accepted, the proposed resolution binds the provider. This is not a court trial with a general invitation to restate the whole commercial relationship. The procedure has an eligibility screen, an application, a response from the provider, a draft proposal, comments and a final proposal. ComReg’s 2024 procedures aim to issue a final proposal within 60 working days of acceptance, provided the other steps meet their timeframes; the regulator may extend the period and inform the parties.

A submission therefore enters a record with a purpose and a clock. The best material helps identify the disputed issue, the evidence that bears on it and the remedy sought. More words are not automatically more evidence. An AI-produced chronology may be useful if each date points back to a message, account record or receipt. It is less useful if a plausible sentence has no source or if legal vocabulary makes an uncertain claim look settled.

Readability and proof are different jobs

ComReg’s examples make the distinction concrete: unverifiable or incomplete references, fabricated facts or citations, incorrect references to legislation or case law, and submissions so long that they delay the process. These are not all the same failure. A wrong citation is a verification problem; an excessive submission is a relevance and review-capacity problem. A document can be concise but false, or long while containing relevant evidence. Human review has to test both content and fit.

The provider is also a party, not simply the subject of a consumer narrative. A provider may hold account, billing, network or customer-contact records that an individual cannot see. The individual may possess the original complaint, screenshots, payment evidence or notes from calls. AI can help either side organise those materials, but it cannot make one side’s account authoritative by making it sound more formal. The evidence must remain linked to its source, and any disagreement must still be decided through the procedure.

The potential access benefit is real as a mechanism, though not measured in the sources reviewed. Someone who is not comfortable writing a formal account may use a tool to turn a sequence of events into clear language or to translate it for review. The same low-cost drafting can also produce a long brief, a fabricated authority or an unsupported allegation. Whether the first effect outweighs the second depends on what the submitter checks and on whether the process can ask for correction without mistaking polished language for proof.

One notice, more than one route

ComReg’s announcement expressly includes postal services, but the postal route should not be collapsed into the electronic-communications procedure. Postal disputes follow separate Procedures, ComReg 24/102a, and users can be senders or addressees. ComReg says the provider’s complaint-code steps must be completed before an application; its guidance asks for clear, chronological material and supporting documents where available. That makes a missing receipt or unclear tracking history a different evidentiary issue from a disputed broadband invoice, even though both may be described in a submission prepared with AI.

The separation matters for providers and business users too. ComReg’s complaint-handling code sets expectations for provider complaint channels, acknowledgement, response and record retention. The formal dispute procedures govern a later stage. AI-generated text does not change which route applies, whether a dispute meets the route’s eligibility conditions, or what documents the party needs to establish a claim.

What the published numbers do not show

ComReg reports 36 domestic formal disputes in 2025, with 19 per cent discontinued and an average resolution time of 62 working days. Its published figures describe a process before the October 2026 AI notice. They do not state how many parties used AI, whether an AI-assisted submission was corrected, or whether AI contributed to the duration of any case. The 62-day figure is also an average from a procedure whose stated target is 60 working days, not proof that AI made a process miss its target.

That evidence gap should limit the claims made about the guidance. ComReg identifies risks it wants parties and case handlers to manage; it does not publish an incident count or a measured productivity result. The notice is an Information Notice, and the sources reviewed do not establish a new AI-specific eligibility test, an approved-tool list or a ban. Nor do they establish that ComReg itself uses AI in adjudication. Those questions require separate evidence.

The operational test is narrower and more useful: can the submitter explain where each material claim came from, distinguish source documents from AI-generated summaries, and revise a file when ComReg asks? A well-run AI workflow should make those checks easier to perform. If it hides the provenance of a statement, it has made the record harder to assess even if the prose is cleaner.

For now, the guidance places the control point where the case enters the institution. AI may help prepare the submission; it does not own the facts, the legal references or the decision. That leaves room for accessibility and better organisation while preserving the basic requirement of a dispute process: each party must be answerable for the material it puts before the decision-maker.

Sources