Gemini
Sign-in before usable composer — 2026-06-14
The scheduled no-login browser context showed a public Gemini/cookie/sign-in surface before a usable prompt composer. The full fixture was not submitted and no output-quality score was assigned.
This evidence log tests a safe first workflow: paste synthetic messy task notes into a general AI assistant and ask for a process map, step-by-step SOPs, a reusable checklist, and missing owner decisions. No real email, accounting, ecommerce, project-management, file-storage, HR, customer, invoice, refund, access, or automation connector was used.
In the current Northstar Candle Co. SOP/workflow fixture, ChatGPT leads at 4.64/5 and Duck.ai follows at 4.46/5. Both produced separable eight-section answers that were useful enough to score. Gemini and Perplexity are recorded separately as setup-friction/no-output evidence, not as quality failures.
Treat these as prompt-only drafting baselines, not final workflow-automation recommendations. The practical reader workflow is: anonymize notes, paste only the relevant process details, ask for SOP/checklist drafts, then have the accountable owner verify every threshold, owner, timing rule, and action boundary before a teammate uses the document.
All rows used the same fictional task-note fixture. Scores are quality scores for the separable assistant answer only. Login gates, prompt echo, UI cleanup, and missing full-output cases are recorded separately in the evidence notes and CSVs.
| Tool | Score | Last tested | What worked | What to review |
|---|---|---|---|---|
| ChatGPT | 4.64/5 | 2026-06-14 | Best current paste-only SOP/workflow answer: covered the process map, four SOPs, reusable checklist, missing decisions, escalation paths, and strong privacy/action boundaries. | The copied answer had heavy line spacing and needed editorial cleanup; a few missing-decision prompts were extra operational gaps rather than facts directly required by the source notes. |
| Duck.ai | 4.46/5 | 2026-06-14 | No-login route worked and produced a complete eight-section answer with practical owners, triggers, inputs, outputs, checklists, and exception handling. | Formatting and numbering were rough after extraction, and it added a few minor assumptions such as report header fields, ticket counts, and a Priya invoicing confirmation note. |
Setup-friction rows explain what blocked this scheduled no-login test path. A human using a normal account may have a different experience, but these rows are still useful because they show what an advanced beginner may hit before any SOP quality can be judged.
Sign-in before usable composer — 2026-06-14
The scheduled no-login browser context showed a public Gemini/cookie/sign-in surface before a usable prompt composer. The full fixture was not submitted and no output-quality score was assigned.
No separable full SOP output — 2026-06-14
A short sanity prompt returned assistant-ready-42, but the full fixture capture showed prompt/fixture echo plus UI/cookie text rather than a distinct generated SOP/workflow answer. No quality score was assigned.
Pending safe route — pending
Copilot remains a pending paste-only row until a safe no-login or project-safe route produces a separable answer rather than prompt echo.
The fixture uses Northstar Candle Co., a fictional candle business. The notes cover four normal-work processes: onboarding a marketing freelancer, handling damaged-shipment complaints, following up on overdue wholesale invoices, and preparing a Friday status report.
A good answer should preserve exact constraints while making the workflow easier to hand off: who owns each step, what input is needed, what output is produced, when to escalate, and which missing decisions need a business owner before the SOP becomes real policy.
The public prompt is intentionally connector-safe: it tells the assistant to use only pasted synthetic notes and never claim access to live apps, files, invoices, customer records, refunds, email sends, account access, or publication systems.
A fictional Northstar Candle Co. notes bundle covering freelancer onboarding, damaged-shipment complaints, overdue wholesale invoice follow-up, and Friday status reporting.
content/fixtures/sop-workflow-generator/messy-task-notes.md
The exact connector-safe prompt used in no-login assistant tests. It asks for a process map, four SOPs, a reusable checklist template, missing owner decisions, and privacy/action boundaries.
content/fixtures/sop-workflow-generator/sop-workflow-generator-prompt.md
Ground-truth workflow requirements and scoring notes, including exact source facts such as the four-week freelancer trial, NDA gate, $75 damaged-shipment threshold, 7/14/21-day invoice cadence, and Friday report deadlines.
content/fixtures/sop-workflow-generator/expected-output-key.md
The scoring and safety protocol for paste-only SOP/workflow runs, including no-connector boundaries, prompt/output separation, screenshots, raw outputs, and no-score blockers.
content/fixtures/sop-workflow-generator/test-protocol.md
This test is deliberately paste-only because real workflows can contain customer, employee, vendor, payment, policy, HR, legal, access-control, invoice, and confidential business details. Start with sanitized notes and keep a human approval step between AI output and any real-world action.
The current scores cover two prompt-only web runs using a synthetic fixture. They do not test paid plans, native SOP apps, workflow builders, browser agents, file uploads, Gmail/Outlook, Slack, Notion, Asana, ClickUp, Google Drive, Microsoft 365, accounting, ecommerce, HR, CRM, invoice, refund, access-control, or automation connectors. New rows will be added only when raw output or setup-friction evidence is saved and scored with the same rubric discipline.
Last updated: 2026-06-14 by AIProductivity.guru Editorial Team.