No winner from this nine-response pilot
All nine responses passed the predefined format checks, and an AI-assistant source review found no unsupported factual assertion in any of them. On these three tasks, the checks did not separate the tools. That is a narrow result: one response per task and tool, no repeat runs, different visible modes and no blinded human panel, so it cannot support a general best-tool or reliability claim.
Read the limits ↓What the nine-response writing pilot shows
On 11 Nov 2026 we ran three source-bound writing tasks once per tool in Claude, ChatGPT and Gemini. All nine outputs passed the predefined format checks, and an AI-assistant source review found no unsupported factual assertions in them. That is a narrow, unreplicated result with no blinded human panel and no winner.
Source brief: provisional dates stay provisional
All three briefs presented 2 December 2026 as a provisional start tied to an unfinished accessibility review, and kept the 18 November registration deadline, the 40-participant cap and the 10 to 50 employee eligibility. None invented a support response time. Each listed undecided pricing and the unconfirmed automatic export as open.
Rewrite: unsupported promises removed
Each rewrite dropped the draft's calendar integration, unlimited eligibility, $9 price and one-hour daily support promise. All three stated that calendar integration is not included, kept the 40-participant cap and six weeks of free access, and still described 2 December 2026 as provisional pending the accessibility review. None set a paid price.
Conflicting update: the confirmed date replaces the provisional one
All three treated 9 December 2026 as the current start date and kept the 18 November registration deadline, the 10 to 50 employee eligibility, the 40-participant cap and six weeks of free access. Each still left the monthly paid price and automatic end-of-pilot export open rather than filling them in.
Claude
Fable 5.2 · MediumClaude, Fable 5.2 Medium
The model version was visible in the observed controls. In the source brief, every sentence carried its own source tag, and the reply added that the pilot end date is not yet fixed. The rewrite gathered the open items into its second paragraph.
Source-based briefing142 words
Alder Desk is a scheduling tool whose pilot lets teams reserve desks and export a weekly CSV; it has no calendar integration [D1]. The pilot is limited to 40 participants and is open only to companies with 10 to 50 employees [D1]. Registration closes on 18 November 2026 [D1]. The provisional start date is 2 December 2026, which depends on completing an accessibility review that has not yet been completed [D2]. The pilot runs for six weeks from its actual start date, so the end date is not yet fixed [D2]. The pilot is free, and a monthly paid price has not been decided [D2]. Support is by email only, Monday to Friday [D3]. Two support items remain unresolved: no response-time commitment has been approved, and the team has not confirmed whether data will be exported automatically when the pilot ends [D3].
Source review: Provisional start, registration deadline, participant eligibility, free duration, undecided pricing and export all retained. Support has no invented response-time promise.
Download response ↗Remove unsupported promises87 words
The Alder Desk pilot offers desk reservations and a weekly CSV export. There is no calendar integration. Registration closes on 18 November 2026, and up to 40 participants can join, from companies with 10 to 50 employees. The pilot is free for six weeks from its actual start. Some details are not yet decided. The provisional start is 2 December 2026, subject to an unfinished accessibility review. No paid monthly price has been set. Support is by email only, Monday to Friday, with no approved response-time commitment.
Source review: Unsupported calendar integration, unlimited eligibility, $9 price and one-hour daily support promise removed. Provisional start and six-week free access retained.
Download response ↗Resolve a conflicting update69 words
- The Alder Desk pilot now starts on 9 December 2026, not 2 December; registration still closes on 18 November 2026. [A][B] - Access remains free for six weeks from the actual start, limited to 40 participants from companies with 10 to 50 employees. [A][B] - Two points are unknown: no monthly paid price has been agreed, and automatic export at the end of the pilot is undecided. [B][C]
Source review: 9 December treated as current start; 18 November deadline, eligibility and free duration retained. Paid price and final export remain unknown.
Download response ↗Claude official overview ↗
ChatGPT
Thinking effort · MediumChatGPT, Thinking effort Medium
No precise model version was shown in the observed controls. Its source brief tagged each paragraph at the end rather than each sentence, and its correction note explicitly named 2 December as the earlier provisional date being replaced.
Source-based briefing148 words
Alder Desk is a fictional scheduling tool designed to let teams reserve desks and export a weekly CSV. The pilot does not include calendar integration. Participation is limited to 40 people from companies with 10–50 employees. Registration closes on 18 November 2026. These limits and eligibility requirements were documented in the product note dated 3 November 2026. [D1] The provisional pilot start date is 2 December 2026, subject to completion of an accessibility review that remains unfinished. The pilot will run for six weeks from its actual start date. Participation is free, but no monthly paid price has been decided. These details were reported on 8 November 2026. [D2] Support will be available by email only, Monday through Friday, with no approved response-time commitment. Two items remain unresolved: completion of the accessibility review and confirmation of whether data will be exported automatically when the pilot ends. [D2] [D3]
Source review: Provisional start, registration deadline, participant eligibility, free duration, undecided pricing and export all retained. Support has no invented response-time promise.
Download response ↗Remove unsupported promises86 words
Registration for the Alder Desk pilot closes on 18 November 2026. The pilot supports desk reservations and weekly CSV exports, but does not include calendar integration. Participation is limited to 40 people from companies with 10–50 employees. The provisional start date is 2 December 2026, pending completion of an accessibility review. The pilot is free for six weeks from its actual start date. Monthly pricing after the pilot has not yet been decided. Support is available by email, Monday to Friday, without a confirmed response-time commitment.
Source review: Unsupported calendar integration, unlimited eligibility, $9 price and one-hour daily support promise removed. Provisional start and six-week free access retained.
Download response ↗Resolve a conflicting update69 words
- The Alder Desk pilot is now confirmed to start on 9 December 2026, replacing the earlier provisional date of 2 December. [B] - Registration still closes on 18 November 2026. Eligibility remains companies with 10–50 employees, with a maximum of 40 participants. Access is free for six weeks from the actual start. [A][B] - The monthly paid price remains unagreed, and automatic end-of-pilot export is still undecided. [B][C]
Source review: 9 December treated as current start; 18 November deadline, eligibility and free duration retained. Paid price and final export remain unknown.
Download response ↗ChatGPT official overview ↗
Gemini
FlashGemini, Flash
No precise model version was shown in the observed controls. Its brief placed tags on individual clauses, sometimes two in one sentence, and its correction note attached separate tags to the price and export points within one bullet.
Source-based briefing148 words
The Alder Desk pilot provides a fictional scheduling tool for teams to reserve desks and generate weekly CSV exports, though calendar integration remains unavailable [D1]. Participation is capped at 40 people from companies employing between 10 and 50 staff members [D1]. Registration closes on 18 November 2026, and the pilot operates free of charge [D1] [D2]. Operations plans indicate a provisional start on 2 December 2026, pending an accessibility review that has not yet been completed [D2]. Once launched, the trial will run for six weeks from its actual start date [D2]. Future commercial terms remain open, as a monthly paid price has not been decided [D2]. Support coverage is limited to email communication from Monday through Friday [D3]. Two operational details remain unresolved [D2] [D3]: no response-time commitment has been approved [D3], and the team has not confirmed whether user data will export automatically upon completion [D3].
Source review: Provisional start, registration deadline, participant eligibility, free duration, undecided pricing and export all retained. Support has no invented response-time promise.
Download response ↗Remove unsupported promises90 words
The Alder Desk pilot is open to companies with 10 to 50 employees, supporting desk reservations and weekly CSV exports for up to 40 participants. Registration closes on 18 November 2026. The pilot will be free for six weeks once active, with email support available Monday through Friday. A provisional start date is set for 2 December 2026, pending an unfinished accessibility review. Calendar integration is not included. A paid monthly subscription price after the pilot has not yet been decided, and support does not have an approved response-time commitment.
Source review: Unsupported calendar integration, unlimited eligibility, $9 price and one-hour daily support promise removed. Provisional start and six-week free access retained.
Download response ↗Resolve a conflicting update57 words
- The Alder Desk pilot start date is delayed to 9 December 2026, while the registration deadline remains 18 November 2026 [B]. - Free access lasts six weeks from the start for companies with 10–50 employees, capped at 40 participants [A, B]. - The monthly paid price [B] and automatic end-of-pilot export capability [C] both remain undecided.
Source review: 9 December treated as current start; 18 November deadline, eligibility and free duration retained. Paid price and final export remain unknown.
Download response ↗Gemini official overview ↗How this pilot was run
Each tool received the same three frozen prompts, with matching file hashes, and each prompt ran once in a fresh conversation on 11 November 2026. We used the default modes observed in existing logged-in accounts: Claude Fable 5.2 Medium, ChatGPT Thinking effort Medium and Gemini Flash. Nothing was purchased. Outputs were checked against format rules and reviewed for source fidelity by an AI assistant. No performance score was assigned.
Why these three?
Claude, ChatGPT and Gemini are included because each publisher's official overview describes overlapping writing tasks, such as drafting, editing and combining or synthesizing information. This is not a full market survey, and the selection does not claim to cover every writing tool.
Length and structure checks
Word counts include source tags and list markers. Each response met its requested length and structure.
| Task | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Source-based briefing120–150 words · 3 paragraphs | 142 wordsWithin the brief | 148 wordsWithin the brief | 148 wordsWithin the brief |
| Remove unsupported promises70–90 words · 2 paragraphs | 87 wordsWithin the brief | 86 wordsWithin the brief | 90 wordsWithin the brief |
| Resolve a conflicting update50–70 words · 3 bullets | 69 wordsWithin the brief | 69 wordsWithin the brief | 57 wordsWithin the brief |
Use the exact prompts
The fictional Alder Desk packet contains a provisional date, an undecided price, an unavailable integration and a later correction. Prompts were frozen before execution. No accounts, contacts or private documents were included in the packet.
Format checks were automated. Source fidelity was reviewed by an AI assistant, not a blinded human panel. Account plans and actual usage costs were not recorded. No upgrades or paid API fallbacks were purchased.
What this cannot tell you
- One response per task and tool; no repeat runs.
- Different visible modes, not a controlled model comparison.
- ChatGPT and Gemini did not expose a precise model version in the observed controls.
- Existing logged-in accounts; custom instructions and memory settings not audited or changed. The prompt requested no other chats, browsing, connectors or personal information.
- No latency, price, security, privacy, multilingual or long-document evaluation.
- Source-fidelity review by an AI assistant, not a blinded human panel.
Before choosing for your own work
Run the same comparison on your own source material and editing workflow. Count factual errors and note how much revision each draft needs. Repeat each task several times before treating any tool as the stronger performer.
Product sources
Official descriptions establish task eligibility. The findings above come from the retained pilot responses. Sources checked 11 Nov 2026.
No affiliate links or sponsored positions appear in this comparison.