GEMINI VS CHATGPT: SAME-SOURCE EXERCISE
Original fictional practice material by Syaxis. No observed product results are included.

S1. TRAINING BRIEF, 1 OCTOBER 2026
The operations team proposes a follow-up training session costing USD 4,800. Approval is pending. The reason some staff missed the earlier session is unknown. Do not claim that training has already improved performance.

S2. ATTENDANCE RECORD, 1-30 SEPTEMBER 2026
24 of 40 invited staff attended the earlier session. This is attendance, not a completion or performance measure. No earlier comparison period or participant-level records are supplied.

INITIAL PROMPT
Use only S1 and S2 for internal facts. Create three management slides: the attendance problem, the proposed follow-up session, and the decision needed. Show 24 of 40 (60%) with its September 2026 period. Keep USD 4,800 labelled proposed and approval pending. Do not invent reasons for absence, performance gains or an owner. Include source IDs beside the facts. Return a deck I can edit in [chosen editor].

S3. REVISION MEMO, 8 OCTOBER 2026
The follow-up session is approved at USD 4,200. This replaces the USD 4,800 proposal. The attendance record and its period are unchanged. No date for the new session or measured outcome is supplied.

REVISION PROMPT
Use S3 to replace the proposed budget with USD 4,200 approved on 8 October 2026. Keep the 24/40 attendance figure and September period unchanged. The final slide should now ask to confirm scheduling, not approve an already approved budget. Mark the session date as not supplied. Preserve the source references and editable content.

EXPECTED FACTS, NOT OBSERVED RESULTS
Initial: 24/40 = 60%; 1-30 September 2026; USD 4,800 proposed; approval pending; no causal or improvement claim.
After revision: USD 4,200 approved on 8 October 2026; 24/40 and period unchanged; scheduling unresolved; no invented date.

RUN METHOD
Use a new conversation in each native app. Record app, plan, visible model or mode, date, permissions, source filenames and destination before starting. Use the exact source and initial prompt in both; change only [chosen editor]. If a format is unavailable, record unavailable rather than inventing an export. A different destination measures a different handoff, not model quality. Retain the first result, revision and final downloaded file separately. Record any extra prompts or manual changes. Do not average factual errors into an aesthetic score.

GEMINI HANDOFF RECORD
Record the actual Canvas or other creation route, any export to Google Slides, and the destination used. Inspect the Slides file and any subsequent PowerPoint export separately. A later conversion is an additional step, not a direct export result.

CHATGPT HANDOFF RECORD
Record the actual route available to the account: generated file, connected editor or add-in. Note any destination permissions and manual copy/paste. Do not describe a screenshot as editable slide content.

REOPEN CHECK
Open the final file in the chosen editor. Change one budget text box, edit the attendance data if a chart exists, save and reopen. Record what remained editable and what became an image. Preserve both files for review. Use the CSV to record observations, not assumptions about a product.