Welcome to another issue of HOW TO // AI. We’re glad you made it!

Early on, AI feels almost magical. It’s fast, articulate, and rarely says “I don’t know.” But as you gain experience—especially by cross-checking outputs between models—you start to see patterns. Some errors are subtle. Others are flat-out wrong. What’s striking is that the confidence level rarely changes.

That’s when I realized LLMs don’t truly self-correct in real time. If you point out a mistake, they’ll often admit their mistake and adjust but only within that conversation. Unfortunately, learning doesn’t persist to next time or to another user. Each new chat starts fresh. Accuracy, then, isn’t guaranteed by the model—it’s shaped by how much structure and discipline you impose through prompting.

Once I accepted that, I stopped assuming AI was “smart” and started telling it to be accurate and honest.

While it can help to use a different model like ChatGPT Thinking, it doesn’t always ensure accuracy and honesty.

One of the most effective changes is explicitly telling the model not to guess. This alone reduces a surprising number of errors. For example, just add this to a prompt-

Another improvement comes from prioritizing correctness over speed. LLMs default to fast completion unless told otherwise.

Hidden assumptions are a major source of error. Forcing them into the open makes mistakes easier to spot.

LLMs can internally sanity-check their own outputs if prompted to do so.

Constraining scope prevents speculative or fringe answers.

Finally, a self-audit at the end encourages restraint.

Hopefully these prompts are very helpful but how do you make these improvements permanent? Most modern AI tools—including ChatGPT and Grok—allow you to set a custom profile, system prompt, or preferences section. This is where you win long-term accuracy. Add a short “accuracy clause” there so it applies to every conversation by default.

For example, include something like: accuracy over speed, no guessing, flag uncertainty, ask clarifying questions. Then layer task-specific prompts on top.

Here’s what it looks like in ChatGPT → Settings → Custom Instructions

And in Grok → Settings → Custom Instructions

📌 Key takeaway: Treat this like setting up spellcheck: do it once, benefit forever. Put the accuracy rules in your AI’s saved settings so the tool is trained to pause, admit uncertainty, and ask follow-ups by default. You’ll spend less time catching mistakes and more time using AI for what it’s best at—drafting, summarizing, brainstorming, and accelerating work without quietly making things up (or flat-out lying to you! 😱)

The Rundown: OpenAI is rolling out targeted advertising in ChatGPT for free users and the $8/month ChatGPT Go tier, while keeping higher paid plans completely ad-free.

Details:

Takeaway: This marks OpenAI’s first major move to monetize the free tier through advertising as inference costs continue to rise. Free and Go users will see more commercialized interactions, while ChatGPT Plus, Pro, Business, and Enterprise subscribers remain unaffected. OpenAI is emphasizing strong privacy commitments by not selling user data and ensuring ads do not influence the AI’s answers.

The Rundown: Google is rolling out "Personal Intelligence" to AI Mode in Google Search, allowing Gemini-powered AI to access users' Gmail and Google Photos (opt-in) to generate highly personalized, context-aware responses.

Details:

Takeaway: This integration gives Google Search a major personalization advantage by leveraging users' own data across Gmail, Photos, and other Google services. It enables more useful, life-specific answers without requiring users to repeatedly provide background context. Google emphasizes strong privacy protections through opt-in consent and limited data usage.

🎵 Beatoven - Generates royalty-free, mood-based music you can use as background audio for videos, podcasts, ads, and more. 

🕵️ FaceCheck ID - Runs AI face recognition search to find where a face appears online (use responsibly and respect privacy/laws). 

📬 Instantly - Scales cold email outreach with lead finding, campaign automation, and sales engagement features. 

🎬 Vidfly AI - Turns text and images into finished videos using a “all-in-one” AI video generator workflow. 

📝 Descript - Edits audio/video like a doc with transcription, clip creation, and podcast/video workflows in one tool. 

🧽 EzRemove - Uses AI to remove watermarks, logos, and text from videos online. 

🔒 Okara - Private AI chat that emphasizes security and uses open-source models without training on your data. 

🎞️ Runway - AI video generation + editing toolkit (text-to-video, image-to-video, and creative controls). 

🔎 Perplexity - AI answer engine that searches the web and returns responses backed by citations you can verify. 

This video breaks down the most important tactics from Google’s prompt engineering course and shows you how to apply them fast without sitting through the full training. You’ll learn Google’s core loop Task, Context, References, Evaluate, Iterate, plus the practical add-ons that actually change your results like persona, format, constraints, multimodal prompting, prompt chaining, agent-style prompts, and meta-prompting so you can get better outputs in fewer tries and start using AI like a real productivity system.

Have a question you want answered? Email [email protected] and you just might be featured in an upcoming issue!

This newsletter is your starter kit for mastering AI with confidence. We keep things simple, show real examples, and focus on quick wins you can repeat.

Stay curious, stick with it, and watch your skills grow week by week!

Until next time :)