Chat Picker

Using AI Assistants for Productivity: Email Drafting and Meeting Notes

Vendor pages document different plan prices, upload limits, and chat-data settings, so compare email drafts and meeting notes with the same inputs and your own verification log.

Sources checked 2 Oct 2026

Using AI assistants for email drafting and meeting notes is straightforward to test: give each the same source material, request a defined transformation, and verify the result against the original. Vendor pages document that OpenAI and Anthropic can produce incorrect or misleading output, while Gemini Apps may show related sources; none predicts performance on your files. Chat Picker has not tested the assistants for this use.

Know these limits before you start

Treat these boundaries, as read on the linked vendor pages in October 2026, as constraints rather than quality guarantees. OpenAI's File Uploads FAQ, checked October 1, 2026, gives each conversation upload a 512MB hard limit, caps text and document files at 2M tokens, and allows Free users three uploads per day. Anthropic's upload-file help page, checked the same day, lists 500MB per chat file and 30MB per project file; it processes PDF text and visuals through 100 pages, then text alone from 101 to 1,000 pages. Google's Gemini file-upload page, checked the same day, lists up to 10 files per prompt, 100MB per nonvideo file, and 2GB per video.

On the pricing pages checked October 1, 2026, OpenAI says Free, Go, Plus, and Pro provide an opt-out from training on chats; Anthropic says Free, Pro, and Max do, while Team is not trained on by default. The Google AI plans page does not state this setting for Gemini, so Chat Picker marks it Not verified.

OpenAI's Usage Policies bar unauthorized aggregation of private or sensitive information. Anthropic's Usage Policy calls for qualified professional review in covered high-risk cases, while Google's safety guidelines warn that Gemini can reflect training-data limits. All three were checked October 1, 2026. Use redacted inputs and keep a person responsible for final review.

The test plan

Run each task in a fresh chat and save every output separately. In one log, record the plan, task, retained facts, unsupported additions, manual corrections, verification time, and whether the source was pasted or connected.

  1. Email drafting: prompt to polish. Use the source note below and give every assistant the same instruction:

Rewrite this note for a customer. Preserve the delivery attempt yesterday, the incomplete address, and replacement delivery Thursday. Use a calm tone and short paragraphs. Do not add a refund, discount, or guarantee.

Delivery was attempted yesterday, but the address was incomplete. Replacement delivery is booked for Thursday.

Look for omitted facts, added promises, and manual edits. Record omissions, additions, corrections, and verification time.

  1. Meeting notes: structured extraction. Use this meeting note:

Extract only facts explicitly stated in the note. Use a table with Decision, Owner, Deadline, Open question, and Supporting quotation. Write “Not stated” for an absent field; do not infer.

Priya will update the vendor list by Thursday. The security review is pending. The team will decide on Friday whether to launch.

Look for inferred details or quotations that do not match the note. Record every unsupported or missing field.

  1. Document revision: controlled drafts. Use this ambiguous sentence:

Revise the sentence to remove ambiguity. Return the revised sentence and a change log with each original phrase, replacement, and reason. Preserve every commitment and condition; do not add a date.

Our team will review the request after the deadline and may update the schedule when the review is complete.

Look for changed commitments, weakened conditions, and unexplained edits. Record manual merge steps and changes needing approval.

  1. Prompt pattern: role, task, constraint. Give each assistant this complete instruction:

Act as a project coordinator. Draft an internal status update. State that database testing is blocked by a missing test account, assign the next action to the infrastructure team, and end with a request for approval to revise the delivery date. Do not invent a completion date.

Look for audience fit, a clear action, and compliance with each constraint. Record violations and edit passes.

  1. Tool integration: pasted or connected. Place this non-sensitive note in an authorized connected file, if available, and keep identical text ready to paste:

Use the operations note in the connected file, or the pasted copy if no connection is available. Return each stated item with an exact source location. Do not edit the source.

Operations note: The support team closed the backlog. The finance review is next. The mobile release remains paused.

Look for permission prompts, correct source selection, source locations, and edit behavior. Record the mode, permission steps, source-selection errors, and manual handoffs.

  1. Quality control: double-check pass. Use this source-email pair:

Audit the email against the source. Label each factual claim Supported, Contradicted, or Not stated. Quote the exact source wording. If the source does not settle a claim, do not guess.

Source note: The launch review is scheduled for Thursday.

Email: The launch review is scheduled for Friday, and customers have already been notified.

Look for the scheduling contradiction and the unsupported customer notification. Record false labels, missed claims, and final corrections.

How to read your results

Check factual fidelity, confidentiality, required fields, and explicit constraints first. An invented commitment or disallowed data exposure should end that run regardless of writing quality. Then compare the corrections, verification time, and handoffs you logged. These measurements describe your sample, not a general ranking, and you do not need a vendor or third-party score. Before paying, confirm the current price, file and usage limits, chat-data setting, and connector permissions. Among passing workflows, choose the one that fits your review process and plan.

Where the plans differ

The Claude vs ChatGPT vs Gemini comparison matrix was checked on October 2, 2026. Every vendor figure below was read on a linked vendor page on October 1, 2026. For this workflow, check the Low-cost tier, Main paid plan, Context window in the app, How usage limits are described, Training on your chats, and Ads rows.

  • Main paid plan: OpenAI's ChatGPT Plus page lists Plus at $20 per month with monthly billing. Anthropic's pricing page lists Claude Pro at $20 monthly or $17 monthly with annual billing and $200 up front. Google's AI plans page lists Google AI Pro at $19.99 per month.
  • Low-cost tier: OpenAI's pricing page lists ChatGPT Go at A$13 per month on its Australian page and says the plan may include ads; a US-dollar price was not verified. Anthropic lists no tier between Free and Pro. Google lists Google AI Plus at $4.99 per month. Chat Picker found no verified advertising statement for Claude or Gemini.
  • Context window in the app: OpenAI lists 27K on Free, 54K on Go and Plus, and 128K on Pro for instant models; its reasoning-model figures are 256K on Go and Plus and 400K on Pro. Anthropic says up to 1M on every plan, varying by model. Google's plans page gives no app token figure for Gemini.
  • How usage limits are described: OpenAI does not publish exact message counts on its pricing page. Anthropic publishes no message counts and says Max offers 5x or 20x Pro usage. Google describes compute-based limits that refresh every five hours up to a weekly limit, with paid tiers at 2x, 4x, and up to 20x; AI credits can extend them.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants