Check before you trust

This issue focuses on three practical questions: How common is AI use? What does a privacy setting actually change? And how can a fake recovery offer target someone after a scam? This is an early-September edition. The reporting below was published before this month or is current help documentation. It is not a recap of events that have not yet happened in September.

A national survey found that 49% of U.S. adults reported ever using AI chatbots. That is a measure of use, not skill.
Turning off model training is different from deleting chat history. Check what your actual setting does.
The FBI warned about fake IC3 recovery approaches, including misleading AI-generated videos. Verify the reporting site independently.

This is the September 2026 sample issue, in full.

Subscribers receive later issues as a PDF in the Stripe email. $8 each month. Cancel from Stripe. Issues already emailed remain yours. Not included in the Household Set.

Subscribe · $8 / month on Stripe Compare products

Use vs expertise

AI use is common. Expertise is different.

Pew Research Center published “Americans and AI 2026: Chatbots, Smart Devices and Views on Impact” on June 17, 2026. Its survey of 5,119 U.S. adults was conducted February 17–23, 2026. It found 49% reported ever using AI chatbots and 24% reported daily use.

Those figures describe U.S. adults overall. They do not describe only people ages 55–75, and they do not measure how well people check an answer. A usage percentage cannot establish whether an assistant will help with your paperwork or whether you should pay for one.

Task: ________________
Useful result: ________________
Corrections needed: ________________
Would I use this method again? Why? ________________

Privacy

Check the setting’s actual job

OpenAI’s Data Controls FAQ explains how to stop ChatGPT conversations from being used for model improvement. In the signed-in web interface, it directs users to the profile menu, Settings, Data Controls, and “Improve the model for everyone.” Turning that off does not remove conversations from chat history.

This is an example from one provider, not a promise that all assistants have the same controls or menu layout. “Does this train the model?” and “Is this conversation saved?” are different questions.

A misleading conclusion to avoid: “I switched off training, so anything I type is safe.” That conclusion does not follow from the setting.

Write one thing the setting changes: ________________
Write one thing it does not establish: ________________
Official help page I checked: ________________
Date checked: ________________

The second scam

Watch for the recovery pitch

The FBI’s July 20, 2026 IC3 warning describes scammers posing as personnel connected with internet-crime complaints. It includes fake recovery offers, spoofed sites, and AI-generated videos. The warning says IC3 does not recover funds through social-media platforms and does not ask for payment to recover lost money.

A fictional example: “Your complaint has been approved. Pay a processing charge and confirm your account details to release recovered funds.” Treat the message as an unverified request. Do not pay or follow its link. Type ic3.gov yourself.

Exercise

Compare a summary with its source

Use these fictional notes; no personal information is needed: The community reading group will meet Saturday at 2 p.m. The room is booked for one hour. Attendance is limited to 12 people. Refreshments have not been confirmed. The organizer will announce the final book choice later.

Ask your assistant: Summarize these notes in three short points. Use only the supplied facts. Keep the attendance limit and anything still undecided. Do not invent refreshments or a book title.

A sound summary should keep Saturday at 2 p.m., the one-hour room booking, the 12-person limit, and the uncertainty about refreshments and the book. Combining facts is fine. Replacing uncertainty with a promise is not.

What the summary got right: ________________
What I corrected: ________________
One task where this check would help me: ________________

Decide

What deserves your attention

Do now: choose one privacy control to understand; practice one independent-verification response; keep one useful prompt from the exercise.

Watch, without buying: if a service you already use adds AI, ask what job it does and whether it changes what data you share.

Skip for now (Waypoint view): you can skip a new app if you cannot name a task it would improve. A dramatic demonstration, popularity claim, or countdown is not a reason by itself to subscribe.

This month’s useful prompt

Summarize these notes in three short points. Use only the supplied facts. Keep the attendance limit and anything still undecided. Do not invent refreshments or a book title.

[paste the fictional notes]

My decision sheet

Choose one privacy control to understand. Practice one independent-verification response. Keep one useful prompt from the exercise. You can do these on separate days.

Watch, without buying. If a service you already use adds AI, ask what job it does and whether it changes what data you share. Read its explanation before enabling access to email, files, contacts, or payments.

Skip for now. You can skip a new app if you cannot name a task it would improve. A dramatic demonstration, popularity claim, or countdown is not a reason by itself to subscribe.

Something I will use this month: ________________
The task it will help with: ________________
Something I will verify: ________________
The official source I will use: ________________
Something I will skip: ________________
Why: ________________

Four terms from this issue

Training: The process of developing or improving a model using examples.
Chat history: Saved conversations visible in a product; retention rules are a separate question to check.
Spoofed site: A site designed to look like another organization or service.
Recovery scam: A new fraud attempt that claims to help recover an earlier loss.

[B1] Pew Research Center. Americans and AI 2026. Published June 17, 2026; survey February 17–23, 2026; 5,119 U.S. adults. Supports the overall 49% ever-use and 24% daily-use figures. It does not measure individual competence. Source

[B2] OpenAI Help Center. Data Controls FAQ. Checked September 8, 2026. Supports the distinction between the model-improvement setting and visible chat history. Source

[B3] FBI Internet Crime Complaint Center. FBI Warns of Scammers Impersonating the IC3. July 20, 2026. Source

[B4] Federal Trade Commission. What To Do if You Were Scammed. June 2026. Source

This issue contains dated reporting, Waypoint interpretation, and original exercises. It does not claim to cover every AI development or to recap the entire month of September. No savings, learning results, or fraud-prevention outcomes are guaranteed. Later issues will note what changed since the previous issue.