Quick answer
This site's 7-day plan, offered as advice and not as a finding: on day 1 pick one chat assistant and set its privacy controls. On days 2 to 6 try 5 tasks you already do. On day 7 count the outputs you kept after review and decide whether to continue. Read every output before use.
Key takeaways
- This is this site's method, not a study result: 1 chat assistant, 5 tasks, 1 check habit, 1 written count. It promises no time saved and no outcome.
- Day 1 is privacy, before any task. OpenAI, Anthropic and Google each document a control over whether chats are used to improve their models, and each treats feedback separately.
- This site's never-paste rule: nothing you would not want a stranger to read.
- All three vendors say their assistant can be wrong, so a person reads and checks everything before it reaches a customer.
- Day 7 rule, this site's: continue a task if you kept the output and only fixed wording; change the task if you fixed a fact; stop if no task ended up kept.
How do you start using AI in your business in the first week?
Spend 7 days on one chat assistant and 5 tasks you already do: rewrite an email, summarize a document, reply to a review, outline a post and draft a customer FAQ, reading every output before it is used. This is this site's method, written as advice. It is not a finding, and it promises no time saved or result.
What do you set up on day 1, before pasting anything?
Choose one assistant, keep it all week, find the vendor's own statement on whether chats are used to improve its models, and set the control that statement names. The table gives the same kinds of statements for each, as read on 7 October 2026.
| OpenAI (ChatGPT) | Anthropic (Claude) | Google (Gemini Apps) | |
|---|---|---|---|
| What the page says about training on individual accounts | "When you use our services for individuals, such as ChatGPT and Codex, we may use your content to train our models." Framed as an opt-out ("To opt out, turn off Improve the model for everyone under Settings > Data controls in ChatGPT"); no default stated for individuals | No default stated on the page we read. Used to improve Claude if "You choose to allow us to use your chats and coding sessions to improve Claude", if conversations are "flagged for safety review", or "By otherwise explicitly opting in to training" | No default stated on the page we read. If Keep Activity is on, "Google uses your activity to provide, develop, and improve its services (including training generative AI models)" |
| The control the page names | "Improve the model for everyone" | "Model Improvement" in Privacy Settings | "Keep Activity" setting |
| No-history mode and training | "Temporary chats are not used to improve OpenAI models while they remain temporary, even if you choose personalization." | "Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings." | "Temporary chats are not used to train Google’s AI models." |
| Retention stated for that mode | "OpenAI may keep a copy of a temporary chat for up to 30 days for safety purposes." | Incognito: "are retained for either 30 days (default), or longer in accordance with your organization's custom data retention setting (available for Enterprise plans)" | Temporary chats, and chats with Keep Activity off, "are retained with your account for 72 hours" |
| Feedback | "If you choose to provide feedback, the entire conversation associated with that feedback may be used to train our models." | "We may use your feedback to analyze the effectiveness of our Services, conduct research, study user behavior, and train our AI models as permitted under applicable laws." | With Keep Activity off, future chats "won’t be used to train our AI models, unless you choose to send Google feedback." |
| Safety or human review | Human review: not stated on the pages we read. Safety: a temporary-chat copy may be kept up to 30 days "for safety purposes" (see the retention row). | Conversations "flagged for safety review" may be used or analyzed "to improve our ability to detect and enforce our Usage Policy, including training models for use by our Safeguards team" | "Even if your Keep Activity setting is off or you use temporary chats, Google still uses your chats to respond to you and help protect Google, our users, and the public, including with help from human reviewers." Also: "Data reviewed by service providers is disconnected from your account and saved for 3 years." |
A page that states more is not shown to do more. "Not stated on the pages we read" means we did not find it, not that it does not happen. Retention periods are for each vendor's own no-history mode, not a ranking.
Account type matters. OpenAI's page also says: "By default, we don’t use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, or our API to improve our models." Anthropic's page is "about our consumer products such as Claude Free, Pro, Max"; Google's says work or school accounts "may be subject to different data handling terms." For a work account, read your vendor's page for that type.
What should you never paste into the assistant?
This site's rule, which is not a legal opinion: paste nothing you would not want a stranger to read. Google's page says: "Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services, including machine-learning technologies." This site applies the same logic to any assistant.
Your never-paste list:
- Customers' names with phone numbers, addresses or account details. Replace them with "Customer A".
- Payment card numbers, bank details, passwords and login links.
- Medical, legal or tax details about a person, and employee records.
- Unsigned contracts or bids, and anything you promised a customer to keep private.
Skip feedback on sensitive chats: each vendor treats feedback separately from the privacy control.
What do you paste on days 2 to 6?
Each day, do one task with text you wrote or can safely share, give the assistant the facts it needs, and run the check before using the result.
| Day | Task | Paste or tell it | Check before use |
|---|---|---|---|
| 2 | Rewrite an email | One email you wrote, and who it is for | Every date, amount and name against your records |
| 3 | Summarize a document | The text of one document you are free to share | Each point against the original wording |
| 4 | Reply to a review | The review text and, from you, the facts of what happened | That the reply promises nothing you will not do |
| 5 | Outline a post | The topic and three points you want to make | That each claim in the outline is one you can support |
| 6 | Draft a customer FAQ | Your real hours, policies and prices, typed by you | Each answer against your own records |
Everything on the never-paste list stays out.
What is the one check to apply to every output?
Read it aloud, then verify every number and name against your own records, before it goes anywhere. This is this site's habit. Each vendor's page carries a caution of its own, worded differently:
- OpenAI: "Use ChatGPT as a first draft, not a final source. Please check important information."
- Anthropic: "Users should not rely on Claude as a singular source of truth and should carefully scrutinize any high-stakes advice given by Claude."
- Google: "Gemini Apps can make mistakes. When using Gemini Apps, double-check responses and don’t rely on responses from Gemini Apps for professional advice."
This site's review rule: nothing reaches a customer, a review page or your website until a named person at your business has read it. In week one, that person is you. For the email and review cases, see AI email responder for small business and AI review response generator.
What do you count on day 7, and how do you decide?
Count three things per task: tried, kept after review, and corrections needed.
| Task | Tried (yes or no) | Kept after review (yes or no) | Corrections (how many, and wording or fact) |
|---|---|---|---|
| Document summary | |||
| Review reply | |||
| Post outline | |||
| Customer FAQ |
A correction of wording is a changed phrase, tone or length. A correction of a fact is a changed number, name, date, policy or claim. "Kept" means you would use the output after your own check. This count measures neither time nor money and sets no benchmark.
This site's rule, which is advice and not a statistical test, for each task:
- Continue if you kept the output and every correction was wording. Do the same task again next week, with the same check.
- Change the task if you discarded the output, or if you corrected a fact, even if you kept the output. Retry with better input or swap the task.
- Stop for now if no task ended in a kept output.
Continue and change are judged per task; stop is judged across all five. A task marked "tried: no" can be carried into next week or dropped; that is this site's suggestion.
What comes after week one?
Next, repeat the tasks marked "continue" before adding anything. If one becomes routine and you want it to run without pasting, see the AI automation guide. For what other US businesses report using AI for, see what businesses use AI for: self-reported Census survey results, not evidence a task works for you.
How was this page made?
This page was drafted with Claude, an AI model made by Anthropic, and checked against each vendor's published material on 7 October 2026. Anthropic is one of the three vendors covered here, alongside OpenAI and Google, and each is described from its own help or privacy page, read in a browser from outside the United States. We did not read work-account pages. The 7-day plan, the never-paste list, the day 7 count and its rule are this site's own method, not a tested result.
FAQ
Do I need to automate anything in the first week?
No. This plan uses one chat window and pasted text. Automation is covered in the live AI automation guide linked on this page; this site suggests waiting until a task has passed the day 7 count.
Which assistant should I pick on day 1?
This page does not rank assistants. Pick one whose privacy sentences you have read and can accept, and stay with it all week so the counts are comparable.
Can I paste customer emails into the assistant?
This site's rule is to paste only what you would be comfortable having a stranger read. If an email contains personal details, remove them first or leave it out.
How do I know after one week whether AI is worth continuing?
Use the day 7 count: tasks tried, outputs kept after review, and corrections needed. This site's rule maps those to continue, change the task or stop. It does not measure time or money.