Three general assistants do most of the writing work
ChatGPT, Claude and Gemini lead. A seat costs about $20 to $30 a month. A detector score is a reason to read closely and is not proof.
What to do about tools
- Pick one general assistant and learn it well. Keep a second for a different opinion on a draft.
- Use the team plan your agency pays for when you handle client material.
- Do not accuse a writer, or clear a text, on a detector score alone.
- Do not pay for a humanizer. It removes the signs a detector reads and adds nothing a reader wants.
ChatGPT, Claude and Gemini lead
| Measure | ChatGPT | Claude | Gemini | Copilot | Source |
|---|---|---|---|---|---|
| Marketers who use it | 81% | 65% | 60% | 26% | Social Media Examiner, July 2026, 681 marketers, 63% in the US |
| Marketers who call it most important | 39% | 42% | 11% | 7% | Same survey |
| Marketers who use AI for content and use it | 75.7% | 83.2% | 57.9% | n/a | Ahrefs, 30 September 2026, 280 AI users among 301 respondents |
Firms pay for more than one. About 79% of Anthropic's business customers also pay for OpenAI (Ramp, February 2026).
Staff in a holding company often reach the same models through an internal platform. WPP says more than 85,000 of its 108,000 people use WPP Open each month, on OpenAI, Google and Anthropic models (Fortune, 21 January 2026).
People edit and summarise with them more than they write from nothing
- About 40% of work-related ChatGPT messages are writing tasks. About two-thirds of those ask the model to edit, critique or translate the user's own text (Chatterji and others, NBER, September 2025).
- 74% of US agencies use generative AI to summarise documents and communications, and 70% use it for research and competitive intelligence (Forrester with the 4A's, June 2026).
- PR professionals who use AI use it for editing (86%), research (76%) and writing (74%) (Muck Rack, vendor survey, January 2026, 564 people).
- 85% of 142 brand and agency professionals say their company uses tools as sold. 40% say it also builds tools on an existing model (Digiday+ Research, January 2026).
What the main tools cost a seat
Read from each vendor's page on 7 October 2026. Check the page before you buy.
| Tool | Individual plan | Team plan | Source |
|---|---|---|---|
| ChatGPT | Plus $20 a month | Business $25 a user a month, or $20 billed yearly | OpenAI, OpenAI help |
| Claude | Pro $20 a month, or $17 billed yearly | Team $25 a seat a month, or $20 billed yearly | Anthropic |
| Gemini | AI Pro $19.99 a month | Google Workspace Business Standard $14 a user a month with a one-year commitment | Google, Workspace |
| Microsoft 365 Copilot | n/a | $30 a user a month in the US, £23.10 before VAT in the UK, paid yearly | Microsoft US, Microsoft UK |
| Jasper | Pro $69 a seat a month, or $59 billed yearly | Custom | Jasper |
| Notion AI | n/a | Business $24 a member a month, or $20 billed yearly | Notion |
Detectors are better than they were and still easy to defeat
- In an independent test, three commercial detectors wrongly flagged 1% or less of human text of medium to long length (Jabarian and Imas, University of Chicago, August 2025).
- In the same test, a commercial "humanizer" raised one detector's miss rate to about 50% or more in most of the cases tested.
- Text from a model that was fine-tuned on an author's work was flagged 3% of the time, against 97% for prompted text (Chakrabarty, Ginsburg and Dhillon, preprint, revised March 2026).
- Seven detectors wrongly flagged 61% of essays by non-native English writers. That test used 2023 versions and we found no repeat (Liang and others, 2023).
- The US regulator found one vendor that advertised 98% accuracy when a test showed 53% (FTC, April 2025).
- OpenAI answers the question "Do AI detectors work?" with "In short, not in our experience" (OpenAI help page).
So a high score on ordinary text is likely to be right, and a low score proves nothing. Use a score as a reason to read the text closely.
31% of UK workers who use AI at work do not tell their employer
That figure and the one in six who pay for a tool themselves are from a survey of 25,000 people (The Next Web on Deloitte and Ipsos, September 2026).
Half of PR professionals now work under a company AI policy, against a fifth in 2024 (Muck Rack, vendor survey). Find yours before you paste a client's document into a tool.