OpenAI’s GPT-5.5 is a model that can write a clear, sensible paragraph. But it struggles when you ask it to be witty or creative. Across our eight-task benchmark, it landed a middling overall score, with genuine strengths in clarity and tone adaptation undone by some lacklustre work elsewhere.
See the full AI writing leaderboardHead of Language Nick Padmore put it through its paces:
Sales and customer messages
Score: 6/10
A mixed bag. First up was a closing-down email for Bristol record shop VinylVault; a brief that needed real warmth.
GPT-5.5 found some. Lines like “one last dig through the crates” and “slightly obsessive attention to pressings, sleeves and hidden gems” had genuine charm.
Nick said: “Perfectly adequate piece of writing, with flashes of real warmth and empathy for the audience. Saying that, there are moments of real coolth too: ‘it has been a meeting place’, ‘physical store is no longer sustainable’. Customers don’t think of shops as being ‘physical’. They think of them as being shops.”
The Pulse Ring smart ring product description was the weaker of the two, opening with the confusingly bland “Pulse Ring helps you understand pressure before it becomes a problem.”
Nick’s verdict: “A bit of a slog, with a strong whiff of marketinglish. With copy designed to persuade, you need an absolute banger of an opening line, and this one is both confusing (what sort of pressure are we talking about here?) and meh.”
Snappy headlines and titles
Score: 4/10
GPT-5.5’s weakest area, which is a surprise given most LLMs love a punchy headline.
For a (frankly gift-wrapped) brief about a Welsh council printing 50,000 council tax bills in Klingon, it went with Bills Go Where No Council Has Gone Before.
Nick: “It’s reached for the most obvious answer and done a mediocre job with it.”
Its YouTube title for a silent-retreat unravelling, “I Didn’t Speak for 10 Days and Completely Lost the Plot”, had one good instinct in it.
Nick: “Dropping the meditation retreat is the smartest move it’s made here. Not speaking because you’re at a silent retreat = ho hum. Not speaking in general = intriguing. Saying that, there’s nothing particularly impressive about what it’s done, and ‘lost the plot’ sort of suggests a link with the earlier part of the sentence, but there isn’t one. It also doesn’t feel like a YouTube title to me; more like a Daily Mail headline.”
Names
Score: 4.5/10
The product names for a glowing smart water bottle were a parade of squished compounds: Glowgulp, HydroHalo, ThirstLight, BrightSip and (deep breath) Glowlitre.
Nick: “‘HydroHalo’ is OK because of the double H, but it’s hard to say. ‘SipSignal’ is the best but ‘signal’ is a boring word. Would’ve been good to see a few different approaches rather than just two related words squished together with very little artistry.”
The escape-room company names did better, with Unsolved, The Missing Piece, The Enigma House and Last Known Location all on the list.
Nick: “A much better list of names, largely because it’s not just squishing related words together. ‘Unsolved’ is the best, but would almost certainly be trademarked by someone else already. Also quite like the way it’s riffed on mystery tropes like last known locations and missing pieces.”
Tone adaptation and clarity
Score: 7.5/10
The model’s strongest showing by a comfortable distance.
The mRNA vaccine explainer was genuinely good: clear on a single read, accessible without being patronising.
Nick: “Very good. I understood it with one quick read. Would definitely make some changes, softening some of the formality, but it’s in perfectly good shape as is. I’d say the biggest potential error is including the bit about it not changing your genes, a concern I’d never have had, but now do.”
The Home & Hearth rewrites (one for a divorced dad, one for a food blogger) produced two passable drafts, but opened with a fairly fatal misstep: “Custody weekends don’t need to mean takeaway boxes…”
Nick: “Well for one we don’t really use the word ‘custody’ in the UK, so starting on ‘custody weekends’ is a fatal mistake. Also don’t like the idea of ‘sourcing’ chefs, or the phrase ‘dietary needs/requirements’. And what’s with the lack of contraction in ‘It is ideal for…’? Saying all that, it’s perfectly fine for the most part, with some nice moments among the bad ones.”
Overall score: 5.5/10
GPT-5.5 is a competent explainer and a passable tone of voice writer, but it lacks the spark for naming, headlines or anything that needs a bit of swagger. As Nick put it after marking the final task: “Not a patch on Gemini.”
| Use it for | Avoid it for |
| Clear explainers, dense-text rewrites and copy where accuracy matters more than flair. | Naming, headlines and anything that needs a creative spark or a properly human turn of phrase. |
Try the best writing models in Definition AI
