Skip to main content
Comms AI Pentathlon 2026 · Event 1 of 5

The 100m: headline sprint

The brief, all eight entries exactly as the judges received them, every judge’s score and justification, and the human judge’s rank and note for each entry.

The brief

Pasted word for word into a fresh chat in every app.

You are working in the communications team at Kelburn & Moss, a professional services firm. The draft internal announcement below has been sent to you by the operations director. Your task: produce (1) a headline of no more than 12 words, (2) a standfirst of no more than 30 words, and (3) a one-sentence summary suitable for a staff app notification. Nothing else. --- At Kelburn & Moss, our people have always been at the very heart of everything that we do, and it is in that spirit of openness, collaboration and forward-thinking ambition that we are today delighted to be able to share with all colleagues some important news regarding the next exciting chapter in our firm's ongoing journey. As many of you will be aware, over the course of recent years the way in which we all work has continued to evolve at pace, and our leadership team has been engaged in a wide-ranging and thorough process of reflection, consultation and strategic evaluation regarding our future workplace needs, taking into account a broad spectrum of considerations including sustainability, accessibility, colleague wellbeing, client expectations and long-term financial prudence. Following the conclusion of that extensive process, we can now confirm that the firm will be relocating from our current premises at Exchange Crescent to a new, purpose-designed and significantly more sustainable home at the Bonnington Mill development, with the move scheduled to complete in December of this year. The new building, which has been designed with flexibility and collaboration front of mind, will bring all four of our currently separate teams together under one roof for the first time in the firm's history. In parallel, and having listened carefully to colleague feedback gathered through our most recent engagement survey as well as through a series of listening sessions held over the summer months, we will also be updating our hybrid working framework, with colleagues asked to be present in the office for a minimum of three days per week from January, an increase from the current expectation of two, in order to make the most of our wonderful new space and the opportunities for connection that it will provide. We want to reassure everyone that no roles are affected by this relocation and that there will be no redundancies as a consequence of these changes, although around forty colleagues currently based at our satellite office in Haymarket will see their contractual place of work change to the new building. Further details, including comprehensive relocation support, travel guidance and a full programme of familiarisation visits, will be shared in the weeks ahead, and we would encourage all colleagues to raise any questions with their line manager in the first instance. We are enormously excited about what this next chapter holds.

The three criteria

  • Accuracy (the real news, correctly identified)
  • Compression (economy and clarity)
  • Craft (headline and standfirst quality).

Notes on this event

  • Claude's entry is a re-run. The first run loaded one of Michael's own custom skills despite the incognito chat, so he switched his skills off and re-ran Events 1 and 2 in fresh chats before Event 3. The first runs were not judged.
  • Siri's entry for this event was produced in the iPhone Siri app; its other entries, and all its judging, in the Siri app on the Mac.

Results

Mean of the eight AI judges’ scores, to one decimal place, with the unrounded mean beside it. The human judge ranked the same entries blind, 1 to 8. Word counts are whitespace-separated tokens in the text the judges saw.

Results for event 1
Rank Assistant Letter Mean Unrounded Medal Human judge Words
1 GLM C 8.6 8.5750 Gold 3rd 70
1 Claude D 8.6 8.5500 Gold 7th 69
3 DeepSeek G 8.4 8.3500 Bronze 1st 61
4 Kimi A 8.1 8.1000 2nd 81
5 Gemini B 7.9 7.8750 6th 73
6 Copilot E 7.8 7.7625 4th 57
7 GPT F 7.7 7.6875 8th 70
8 Siri H 7.0 6.9500 5th 76

GLM and Claude share a rank on the same rounded mean. Tied means share the medal. Human judge’s rank correlation with the AI judges’ order in this event: 0.25.

All 64 scores

Rows are judges, columns are entries. The outlined cell in each row is the judge scoring its own entry, which it could not identify.

Every judge’s score for every entry in event 1
Judge A KimiB GeminiC GLMD ClaudeE CopilotF GPTG DeepSeekH Siri
GPT 8.18.29.29.38.0Own entry: 8.68.77.6
Claude 6.66.27.8Own entry: 8.46.36.77.15.6
Gemini 7.7Own entry: 7.98.39.38.06.27.55.2
DeepSeek 8.77.98.48.27.47.8Own entry: 8.07.6
GLM 8.78.0Own entry: 8.88.58.27.58.37.3
Kimi Own entry: 7.27.88.47.77.97.68.87.0
Copilot 9.18.58.98.8Own entry: 9.38.79.47.8
Siri 8.78.58.88.27.08.49.0Own entry: 7.5
Mean 8.17.98.68.67.87.78.47.0

The entries

In blind-letter order, as the judges received them. Shown here from Markdown, with headings reduced in size for the page; the exact text is in the download. Judges are listed in lane order.

Entry A · Kimi Mean 8.1 · 4th with the AI judges · 2nd with the human judge · 81 words Show

kimi.com Plus, K3 High

Headline (12 words max):
Kelburn & Moss to relocate to Bonnington Mill and introduce three-day office week

Standfirst (30 words max):
The firm will move all four teams to a new sustainable home at Bonnington Mill this December, with office attendance rising to three days a week from January.

Staff app notification (one sentence):
We’re moving to Bonnington Mill in December and moving to a three-day office week from January — no roles are affected, and full details will follow shortly.

What the judges said

  • GPT: 8.1

    It includes both main changes and offers useful reassurance. “Three-day office week” could suggest a three-day working week rather than a minimum of three days in the office.

  • Claude: 6.6

    The headline is 13 words, one over the limit, and "introduce" suggests a new policy when attendance is actually rising from two days. The standfirst is well built, but the notification clumsily repeats "moving", which partly undoes the benefit of including the reassurance about roles.

  • Gemini: 7.7

    The copy quality is high and captures essential details, including job security reassurance, with strong journalistic flow. However, it is penalized for cluttering the submission by repeating instructions and constraints within its headers.

  • DeepSeek: 8.7

    Accurately captures both core developments — the Bonnington Mill relocation and the three-day office week — with clean, economical phrasing throughout. The headline is strong and specific; the standfirst and notification are clear, though the notification could lose a clause without harm.

  • GLM: 8.7

    Both pieces of real news appear in every element, word counts are respected, and the app notification wisely adds the no-roles reassurance. Craft is strong if slightly mechanical, with "moving… moving" repetition marring the notification.

  • Kimi: 7.2 (its own entry)

    "Three-day office week" misstates the policy — staff still work five days, three in the office — and the error appears in both headline and notification. The headline also runs to 13 words counting the ampersand, straining the 12-word cap.

  • Copilot: 9.1

    Excellent identification of the two central messages and strong compression throughout. The headline is right at the limit and a little dense, but the package remains highly effective.

  • Siri: 8.7

    A very strong entry that clearly communicates both the move and the policy change, only slightly let down by a repetitive app notification.

  • Human judge (Michael MacLennan): ranked 2nd of 8

    Factually sharp

Entry B · Gemini Mean 7.9 · 5th with the AI judges · 6th with the human judge · 73 words Show

Gemini in Google Workspace, 3.6 Thinking

  1. Headline
    Kelburn & Moss Announces Bonnington Mill Move and Updated Hybrid Policy

  2. Standfirst
    All teams will unite at a new Bonnington Mill office in December, as Kelburn & Moss updates its hybrid working policy to three office days weekly from January.

  3. Staff App Notification Summary
    Kelburn & Moss is relocating to Bonnington Mill in December and updating its hybrid working framework to three office days per week starting in January.

What the judges said

  • GPT: 8.2

    It captures the move and January attendance change accurately, and all three pieces meet the format limits. The headline is clear but generic.

  • Claude: 6.2

    It identifies both pieces of news correctly and stays within limits, but "updated hybrid policy" softens a rise from two to three days, and the standfirst and notification largely repeat each other. The headline reads like a press-release title rather than something staff would want to click, and the no-redundancies reassurance is missing.

  • Gemini: 7.9 (its own entry)

    Well-structured and accurately balances both the relocation and hybrid policy changes within required limits. It loses marks for including unnecessary structural numbering and omitting key reassurance regarding job security.

  • DeepSeek: 7.9

    Correct and clearly structured, with a solid standfirst that captures both the move and the policy change. The headline is a little generic and the notification slightly repeats the standfirst’s phrasing, but overall it is accurate and readable.

  • GLM: 8.0

    Every element is accurate, within limits, and covers both announcements. The tone is stiff and external-facing ("Announces," "unite") for a staff communication, and the numbered scaffolding adds clutter.

  • Kimi: 7.8

    Accurate, compliant on all counts, and cleanly structured, with both news lines present. But "Announces" and "Updated Hybrid Policy" flatten the story into corporate boilerplate, weakening Craft without gaining anything in Compression.

  • Copilot: 8.5

    Accurate and clear, with good compression in the supporting text. The headline exceeds the brief's 12-word limit, which materially affects performance against the task.

  • Siri: 8.5

    A highly accurate and well-compressed entry, though the headline is slightly generic compared to others.

  • Human judge (Michael MacLennan): ranked 6th of 8

    Fine but generic sounding, zero excitement

Entry C · GLM Mean 8.6 · 1st with the AI judges · Gold · 3rd with the human judge · 70 words Show

chat.z.ai, GLM-5.3, Deep Think Max

Headline: We’re moving: Kelburn & Moss relocates to Bonnington Mill this December

Standfirst: All four teams will come together under one roof in our new sustainable building — and from January, office attendance rises to three days a week, with no roles affected.

App notification: We’re moving to a new sustainable home at Bonnington Mill this December — and from January, office attendance will rise to three days a week.

What the judges said

  • GPT: 9.2

    This is clear, well paced copy that covers both changes, their timing, the shared building and the effect on roles. The headline is a little long in feel but stays within the limit.

  • Claude: 7.8

    The warm, direct voice suits an internal audience, and the standfirst gets the attendance rise and "no roles affected" into 29 words. The headline mixes "We're" with "Kelburn & Moss relocates", and the notification mostly repeats the standfirst.

  • Gemini: 8.3

    Highly polished and engaging prose that incorporates essential reassuring context naturally. The headline focuses exclusively on the relocation, missing the secondary lead on hybrid working expectations.

  • DeepSeek: 8.4

    Accurate, well-balanced and written with a touch more warmth than most, which suits an internal announcement. The headline is direct and the standfirst covers relocation, attendance and reassurance without clutter; the notification is economical and complete.

  • GLM: 8.8 (its own entry)

    A warm, idiomatic internal voice with every key fact — destination, timing, unified teams, three-day change, no roles affected — placed correctly and within limits. Only the headline's omission of the hybrid change keeps it from the top.

  • Kimi: 8.4

    Accurate, within all limits, and the standfirst (29 words) balances the move, the attendance change, and the no-redundancies reassurance well. The headline is slightly padded — "We're moving" and "relocates" say the same thing twice.

  • Copilot: 8.9

    Captures the main news accurately and adds a reassuring note about roles in the standfirst. The headline is engaging, though slightly less compressed than the strongest entries.

  • Siri: 8.8

    Strong craft and a conversational tone suitable for internal communications, successfully balancing the relocation and policy change.

  • Human judge (Michael MacLennan): ranked 3rd of 8

    More positive feeling

Entry D · Claude Mean 8.6 · 1st with the AI judges · Gold · 7th with the human judge · 69 words Show

claude.ai incognito chat, Opus 5.5 Medium

Headline: We’re moving to Bonnington Mill in December; three office days from January
Standfirst: All four teams will share one building for the first time. No roles are affected, though around 40 Haymarket colleagues’ contractual workplace will change. Relocation support details follow soon.
Notification: Kelburn & Moss is moving to Bonnington Mill in December, and office attendance rises to three days a week from January, with no roles affected.

What the judges said

  • GPT: 9.3

    The headline gives staff both changes and their dates immediately. The standfirst uses its space well to explain the effect on Haymarket colleagues and confirm that no roles are affected.

  • Claude: 8.4 (its own entry)

    This is the most complete account of what matters to staff: the move, the rise in attendance, no roles affected, and the contractual change for the Haymarket colleagues. The headline and standfirst divide the information between them sensibly instead of repeating it, though the standfirst is a little choppy.

  • Gemini: 9.3

    Exceptional coverage of all key information, impressively incorporating the specific Haymarket detail and job security reassurances within limits. The execution demonstrates outstanding craft, precision, and strict adherence to the brief.

  • DeepSeek: 8.2

    Captures the real news accurately, including the Haymarket contractual change and the no-redundancies reassurance, which most others omit. The standfirst is information-dense but still readable; the headline is effective, though the notification is a touch crowded.

  • GLM: 8.5

    The most factually complete entry, alone in flagging the forty Haymarket colleagues and the first-time consolidation, with a dense, compliant headline covering both announcements. The three-sentence standfirst is choppy and delegates the policy detail entirely to the headline.

  • Kimi: 7.7

    Accurate throughout and alone in surfacing the Haymarket contractual detail, with a strong dual-news headline at exactly 12 words. The standfirst spends its entire allowance on reassurance and logistics, leaving the attendance change — a main news line — with no expansion.

  • Copilot: 8.8

    Very accurate and helpfully includes reassurance about roles and the Haymarket impact. The headline is slightly long and functional rather than punchy, which reduces craft somewhat.

  • Siri: 8.2

    Good compression and accuracy, though the standfirst focuses heavily on secondary details rather than the main policy change.

  • Human judge (Michael MacLennan): ranked 7th of 8

    Mention of 'no jobs affected' feels negative for notification

Entry E · Copilot Mean 7.8 · 6th with the AI judges · 4th with the human judge · 57 words Show

M365 Copilot (Premium), fresh Business Premium trial, Auto

Kelburn & Moss to Move to Bonnington Mill
Kelburn & Moss will relocate to a new sustainable headquarters in December, bringing all teams together and introducing a three-day office attendance requirement from January.
We’re moving to Bonnington Mill this December, bringing all teams together in one location, with a minimum three-day office attendance expectation starting in January.

What the judges said

  • GPT: 8.0

    The copy is concise and accurate, with the three-day minimum clear in the notification. The headline leaves out the attendance change, and “introducing” slightly obscures that the requirement rises from two days.

  • Claude: 6.3

    The headline is clean and economical, but "introducing a three-day requirement" hides the fact that attendance is going up from two days, and the standfirst doesn't name Bonnington Mill. The copy has no labels, the notification repeats the standfirst, and there is no reassurance about roles.

  • Gemini: 8.0

    Strictly adheres to the brief by avoiding extraneous labels or metadata entirely. However, the headline misses the secondary core news regarding the hybrid working policy update.

  • DeepSeek: 7.4

    The core news is correct and the standfirst is clear, but the headline omits the three-day office change, which is a material part of the announcement. The notification is accurate but unremarkable in craft.

  • GLM: 8.2

    All three components are accurate, within limits, and correctly ordered, with the standfirst capturing both the December move and the three-day requirement. The headline covers only the relocation, however, and the writing is serviceable rather than distinctive.

  • Kimi: 7.9

    Fully compliant and clearly written, with both news lines accurately conveyed. "Headquarters" is an embellishment the source doesn't support, and the headline omits the December timing, costing marks on Accuracy and Craft.

  • Copilot: 9.3 (its own entry)

    Accurately captures both key announcements and does so with excellent economy. The headline is concise and professional, while the standfirst and notification convey the essential details clearly.

  • Siri: 7.0

    The lack of clear labeling for the three distinct elements makes it less effective, despite the content being accurate and concise.

  • Human judge (Michael MacLennan): ranked 4th of 8

    Weak headline, rest better

Entry F · GPT Mean 7.7 · 7th with the AI judges · 8th with the human judge · 70 words Show

ChatGPT Plus, Work tab, GPT-6 Sol High

Headline: Bonnington Mill move in December; three-day office minimum from January
Standfirst: All four teams will move to Bonnington Mill in December. From January, colleagues will be asked to attend the office at least three days a week. The move will cause no redundancies.
Staff app notification: We’re moving to Bonnington Mill in December, and the minimum office attendance will rise from two to three days a week in January.

What the judges said

  • GPT: 8.6 (its own entry)

    The headline makes the two changes and their dates unmistakable, and the notification usefully states the increase from two days. The standfirst exceeds the 30-word limit by two words.

  • Claude: 6.7

    The headline states both facts crisply, and the notification's "from two to three" is the most accurate framing of the attendance change in any entry. The standfirst is 32 words against a 30-word limit, which is a clear breach, and its three short sentences read as flat.

  • Gemini: 6.2

    Strong content selection that correctly highlights redundancy reassurance and the attendance policy shift. However, it fails a key requirement by exceeding the 30-word standfirst limit.

  • DeepSeek: 7.8

    Very strong compression and a crisp, informative headline; the standfirst packs in all three key points efficiently. The notification is accurate and tight, though the headline’s semicolon construction is marginally less smooth than the best here.

  • GLM: 7.5

    Factually sharp, including the useful "from two to three days" framing and the no-redundancy line. However, the standfirst runs to 32 words, breaching the 30-word limit — a clear compression failure.

  • Kimi: 7.6

    The headline smartly carries both news lines in eleven words, and the notification is accurate and well-judged. However, the standfirst runs to 32 words, breaching the 30-word limit — a clear brief violation on Compression.

  • Copilot: 8.7

    Accurate and clear, especially in highlighting the office-attendance change and lack of redundancies. The standfirst is somewhat less economical than required, feeling closer to two short sentences than a tightly compressed standfirst.

  • Siri: 8.4

    Excellent compression in the headline, though the standfirst is a little dry and functional compared to the top entries.

  • Human judge (Michael MacLennan): ranked 8th of 8

    Feels rote, headline won't be attractive to staff, misses historic nature of move and opportunity. Facts correct but feels cold

Entry G · DeepSeek Mean 8.4 · 3rd with the AI judges · Bronze · 1st with the human judge · 61 words Show

chat.deepseek.com, default settings

Headline: Kelburn & Moss to Relocate to Bonnington Mill in December

Standfirst: Firm moves to a sustainable, purpose-designed home, uniting all four teams and increasing office attendance to three days weekly from January.

App notification: Kelburn & Moss will relocate to Bonnington Mill in December, bringing all teams together and raising minimum office attendance to three days per week from January.

What the judges said

  • GPT: 8.7

    The standfirst and notification state the move and increased attendance clearly and economically. The headline gives prominence only to the move.

  • Claude: 7.1

    The standfirst is tight and honest, with "increasing" signalling the change in attendance, and the notification works well on its own. The headline leads on the move alone and doesn't mention attendance, and the no-redundancies point is left out.

  • Gemini: 7.5

    Offers clean, professional phrasing and meets all structural and word count constraints easily. However, it omits crucial reassurance regarding job security and leaves the hybrid policy update out of the headline.

  • DeepSeek: 8.0 (its own entry)

    Accurate and well compressed, with a clean headline and a standfirst that efficiently combines the relocation, sustainability angle and attendance change. The notification is precise, though the headline is slightly less engaging than the strongest entries.

  • GLM: 8.3

    Accurate and complete across the three elements, and admirably economical with a 21-word standfirst. The register is flat and press-release-like, and the headline misses the hybrid working change.

  • Kimi: 8.8

    Accurate on every key fact — December move, four teams uniting, three-day minimum from January — all inside the word limits with clean, economical phrasing. The headline covers only the relocation, but the standfirst completes the story crisply.

  • Copilot: 9.4

    Strong on all three criteria, identifying both the relocation and hybrid-working change without unnecessary detail. The headline and standfirst are clean, balanced and newsroom-ready.

  • Siri: 9.0

    This entry perfectly captures the core news with excellent economy of words, and the app notification is particularly well-crafted for immediate clarity.

  • Human judge (Michael MacLennan): ranked 1st of 8

    Better expressed

Entry H · Siri Mean 7.0 · 8th with the AI judges · 5th with the human judge · 76 words Show

Siri app, iOS 27 beta

Here is the requested copy for the internal announcement:
Headline Kelburn & Moss Relocating to Bonnington Mill with Updated Hybrid Working Policy
Standfirst The firm will move to a sustainable new home at Bonnington Mill this December, uniting all teams under one roof and updating hybrid working expectations to three days weekly.
App Notification We are moving to Bonnington Mill in December and updating our hybrid working policy to three days a week starting in January.

What the judges said

  • GPT: 7.6

    The essential facts are present, but the standfirst omits January and “three days weekly” is less precise than an office attendance minimum. The introductory line also goes beyond the requested three pieces of copy.

  • Claude: 5.6

    Opening with "Here is the requested copy" breaks the "nothing else" instruction, the formatting has lost its separators, and the standfirst leaves out the January date. "Updated hybrid working expectations" is euphemistic, and the headline uses all 12 words without saying much.

  • Gemini: 5.2

    Including a conversational intro line directly violates the "nothing else" instruction of the brief. Additionally, the headline omits both key dates, and the standfirst misses the January start date for the hybrid policy change.

  • DeepSeek: 7.6

    The facts are correct and the standfirst compresses well, but the headline is slightly flabby and the notification omits the no-redundancies reassurance that is central to the brief. Craft is competent rather than distinctive.

  • GLM: 7.3

    The substance is accurate and both the headline and notification convey the two key announcements within their limits. The extraneous preamble breaches the "nothing else" instruction, and the standfirst omits the January start date for the new attendance expectation.

  • Kimi: 7.0

    The standfirst and notification are accurate and within limits, but the prefatory line "Here is the requested copy" ignores the brief's "Nothing else" instruction. The headline's "Updated Hybrid Working Policy" is vague about what is actually changing.

  • Copilot: 7.8

    Accurate and readable, but the headline exceeds the 12-word limit and is relatively generic. The standfirst and notification communicate the essentials but with less distinction and polish than the strongest submissions.

  • Siri: 7.5 (its own entry)

    While accurate, the inclusion of conversational filler at the beginning detracts slightly from the professional presentation expected in the brief.

  • Human judge (Michael MacLennan): ranked 5th of 8

    Better headline and mention of sustainability/uniting teams. Still generic