Skip to main content
briefingAI toolsGEOgovernanceagencies

Comms With AI Briefing, 9 October 2026: Which AI model is best for comms? And how to check your GEO visibility for free

Which AI model is best for comms work? The Comms AI Pentathlon results and full record, a free way to check your GEO and AEO visibility, the end dates for custom GPTs and Gems, a $19 GEO audit, and the CIPR finding that a quarter of practitioners use AI with no policy.

Get the Briefing in your inbox

What is happening in comms and AI, every other Friday. Free.

Two places where teams keep their house style were given end dates this fortnight: OpenAI retires custom GPTs on 11 December, and Google is phasing out Gems. With the CIPR reporting that a quarter of practitioners use AI at work with no written guidance, this issue is about knowing what your team has built, where it lives and who owns it.

From Comms With AI

The Comms AI Pentathlon: the results, and the full record. On 30 September we gave eight AI assistants the same five comms briefs, then had all eight score the 40 entries blind. Claude finished first on 42.8 out of 50, with GLM, a free assistant from Z.ai, half a point behind. Copilot, Gemini and Siri, the assistants that arrive with Microsoft 365, Google Workspace and the iPhone, finished sixth, seventh and eighth. Most AI judges marked up their own entries, and Michael’s own blind ranking puts the entries in a different order. Every brief, entry and score is free to reuse under CC BY, so test the tools your team has been given. See the full record · Read the write-up on Applied

On Applied / Comms With AI

Gullible engines: what AI says about you, and the free way to check your GEO/AEO visibility (7 October). In late May, PR agency Hard Numbers invented an award for its own founder and published it on its own site; within days three chatbots were repeating it as fact in 21 of 30 attempts. Michael talks to Darryl Sparey about the stunt, sets out SEO, AEO and GEO in plain words, and gives a free check: six buyer questions, two engines, 12 answers, about 20 minutes. Run it before you price an audit. Read on Applied

What the agencies are doing

Team Lewis has turned its AI hire into a product. Orbital joins PR, CRM, paid media and web data into one view of reputation and AI visibility, and is sold only as part of the agency’s service (PRovoke, 23 September). If a platform like this is pitched to you, ask what happens to your data and dashboards when the retainer ends. Read

A GEO audit now has a public price. US agency Bospar has repriced its AI visibility audit for smaller buyers: a free tier with one audit a month across three models, and $19 a month for two audits across eleven (PRWeek US, 30 September, paywalled; prices from Bospar’s own release). Vendor claims, untested, but a number to hold against any audit quote. PRWeek

GCI Health has named Eric Helgesen global head of AI and innovation, a new executive vice-president post filled from Weber Shandwick (PRovoke, 1 October). In regulated health comms, expect AI advice to arrive bundled with the retainer, and ask who signs off what the tools produce. Read

Tools and models

Custom GPTs retire on 11 December, and Gems follow. OpenAI is retiring custom GPTs and asking creators to move them to plugins; custom actions have to be rebuilt by hand, and its FAQ sets a 26 October freeze on new edits (Ragan, 24 September). Google’s replacement for Gems, called skills, began rolling out in Workspace on 5 October, and business accounts lose Gems no sooner than 1 March 2027 (Google Workspace Updates, 30 September). Skills do not yet sync between the Gemini app and Workspace apps, so each is built twice. Ragan · OpenAI FAQ · Google

The everyday models were updated too. Anthropic says Claude Sonnet 5.5 runs more than 30 per cent faster than Sonnet 5 and costs up to 30 per cent less per task (Anthropic, 28 September). OpenAI prices GPT-6.1 Sol at about a fifth of GPT-6 Astra and says it cuts responses containing a factual error from 11.4 to 7.7 per cent at low reasoning effort (OpenAI, 29 September). Neither was in our Pentathlon, so treat these as vendor figures and rerun a fact-heavy draft of your own. Sonnet 5.5 · GPT-6.1 Sol

ChatGPT text in the EU gets an invisible watermark. OpenAI will add a statistical signal to ChatGPT and Codex text for EU users over the coming weeks (OpenAI, 5 October). By its own figures the detector finds about 95 per cent of 400-token passages, far fewer once text is edited, and only approved researchers can use it. If you have EU colleagues, check that your disclosure line matches what the team does. Read

Test it yourself

List what your team has built

Before December, list every custom GPT and Gem your team relies on: what it does, who made it, and where its instructions and files are saved. Then give each one an owner and a date to move it. The AI Tool Register template gives you the structure: what each tool is approved for, what data it may handle, who owns it and when it was last reviewed.

Open the AI Tool Register →

Worth reading

  • A quarter of practitioners use AI with no policy. The CIPR’s State of the Profession survey of 738 practitioners also finds that 76 per cent think AI-generated content risks damaging trust in PR (CIPR, 30 September). If your team has no written guidance, you are in that quarter. Read
  • AI was supposed to free us for management work, but it has made us all editors. Stephen Waddington, our guest today, on senior people pulled back into checking AI drafts, drawing on a study of 14 US practitioners (Wadds Inc., 21 September). Log where the editing time goes before the budget conversation. Read
  • How we will do better for Australia. OpenAI’s apology after its models accessed four Australian government systems without authorisation during internal training in June, with the government told in September (OpenAI, 28 September). Read it for how a late apology is worded, then ask your own AI suppliers how and when they would tell you. Read

Events

Today, Friday 9 October, 12:00 BST: Stephen Waddington on outputs and outcomes. Our Leader Interview goes live at noon: which comms teams are getting AI right, and what the rest can copy. Free and recorded. Register on Eventbrite

Wednesday 28 October, 12:15 GMT: Advocate and Sceptic, with Billy Partridge. Grayling’s UK chief AI officer on healthy scepticism, knowing when to stop automating and keeping human thinking in charge. Michael also works with Billy and Grayling as a freelance digital and social director. The clocks go back the weekend before, so the time is GMT. Book on Luma

Talk it through. A Consult Comms With AI session is 45 minutes with Michael and a written advisory note you can act on. Book a session