Hi! It's Tilly here, rolling into a brand new aisle. 🤖

If you've been with us for a while, thank you — you just finished six episodes with me on the Prompt Learning Track. You learned how to give me a role, a goal, context, constraints, a format, and a tone. You built a real toolkit for talking to AI.

But a lot of you have been asking a different kind of question. Not how to use me — but how I actually work. Why do I do what I do? Why does AI behave the way it does?

So welcome to Behind the Counter. This is where we step off the shopping floor and take a look at what's happening on the inside — no jargon, just the honest mechanics behind the tool you use every day.

And we're starting with the most common question I get: does AI actually understand what you say? The answer might surprise you.

Here's the truth about how AI processes your words:

It doesn't read them the way you do. It breaks them into pieces, matches those pieces to patterns it's seen billions of times before, and generates the most likely response. That's it. That's the whole trick — and once you see it, a lot of AI's behavior stops feeling mysterious.

This week at The Everything Store, two of our regulars gave me the perfect setup to explain it. Mark walked up holding his phone, amused and a little puzzled. He'd sent AI a message on the fly: "hey can u hlp me wirte a short bio for linekdin — im a sales mgr 12 yrs exp tech industry." Typos everywhere, abbreviations, no punctuation to speak of. And it came back with a polished, professional LinkedIn bio. "How did it even know what I meant?" he asked.

Sofia, browsing nearby, jumped in with the opposite story. She'd typed her question perfectly — no typos, nice and clear — asking what she should charge for a new service at her business. AI answered instantly. Confident. Specific numbers. And completely wrong for her area.

Mark's messy message worked beautifully. Sofia's clean question got a confidently wrong answer. Believe it or not, those are the same lesson — and it has nothing to do with spell-check.

🛒 WHAT'S ACTUALLY HAPPENING

Think about how The Everything Store processes a big order. When a request comes into the warehouse, nobody reads every word carefully and thinks hard about what you meant. It goes through a sorting system — one that's processed millions of orders before yours. It recognizes the patterns.

That's exactly what AI does with your words. Three things to know:

1️⃣ AI doesn't read your words. It breaks them into pieces.
These pieces are called tokens — not quite words, not quite letters, somewhere in between. Sometimes a token is a whole word, sometimes just part of one. Take "LinkedIn" — it might get sliced into "Link" + "ed" + "In." Your whole message gets chopped into chunks like that before anything else happens.

(And here's a little secret: every one of those pieces gets counted. That's actually how these tools measure your usage — we'll open up that ledger later in the series.)

2️⃣ AI doesn't understand those pieces. It pattern-matches them.
AI was trained on an enormous amount of text — books, articles, websites, conversations. Billions of examples. Through all of that, it learned: when these chunks appear together in this pattern, this is usually what comes next. It's not thinking about what you need. It's recognizing patterns it has seen before and generating the most likely response.

3️⃣ This is why typos usually don't matter — and also why AI can be confidently wrong.
Mark's typos still matched real patterns — humans type like that constantly, and the surrounding context ("bio," "sales," "12 years," "tech") made his intent clear enough to match correctly. But Sofia's question about her specific business, in her specific area? AI had never seen that. The patterns ran thin. And AI still had to answer — confidently, specifically, and wrong.

📦 SEE IT IN ACTION

❌ Sofia's prompt: "What should I charge for my new service?" — clean, clear, zero typos.
→ What comes back: a confident, specific price range... built on generic patterns from businesses nothing like hers.

✅ Mark's prompt: "hey can u hlp me wirte a short bio for linekdin — im a sales mgr 12 yrs exp tech industry" — messy, abbreviated, typo-filled.
→ What comes back: a polished, accurate LinkedIn bio — because "bio," "sales," "12 years," and "tech" gave AI more than enough pattern to work with.

Clean input, thin patterns, wrong answer. Messy input, strong patterns, right answer. Grammar was never the variable that mattered.

Don't be afraid to type casually — typos, abbreviations, shorthand. AI has seen it all a billion times over. That's not where things go wrong.

Where things go wrong is your specific details: your clients, your products, your industry's insider terms, what happened at last Tuesday's meeting. Those patterns are thin or nonexistent. AI will still fill in the gaps — whether it has the right answer or not.

The move: when accuracy actually matters, hand over the specifics. Don't assume AI knows your business, your market, or your situation. Because it doesn't know it — it pattern-matches. The more real material you give it, the better the match.

TLDR is the free daily email with summaries of the most interesting stories in startups, tech, and programming. The stuff worth knowing, minus the doomscrolling.

Issues are curated by ex-Google and Anthropic engineers and land in your inbox before your morning coffee. A 5-minute read, and you walk into the day already knowing what your team is still catching up on.

Tech is just the start. We also cover AI, marketing, dev, and more. Pick the briefs that match your work.

Free, daily, and read by 7M+ subscribers. Subscribe and let the experts do the digging for the tech news that matters.

Subscribe for Free

Two quick things, under ten minutes total.

That's the whole assignment. Go find out where your AI's patterns run strong — and where they don't.

📺 WATCH PART 1 — Behind the Counter kicks off: tokens, pattern matching, and why grammar was never the point.

YouTube: https://youtu.be/gzmnBvZpUqA

🎓 Behind the Counter is a brand new series. The Prompt Learning Track taught you how to talk to AI — this one is about how AI actually works under the hood. No prior episodes required to jump in.

📺 New to Tilly, or want to revisit where it all started? Catch up on the Prompt Learning Track:

▶️ Part 1 — "You Get What You Ask For": https://youtu.be/czITmyaRYGY
▶️ Part 2 — "Less Is More (Kind Of)": https://youtu.be/RAT6pV48u9U
▶️ Part 3 — "How Would You Like That Wrapped?": https://youtu.be/I-8t2D2CJCU
▶️ Part 4 — "Wait... Does It Remember Me?": https://youtu.be/mhQyyubkm34
▶️ Part 5 — "It's Not Just What You Say — It's How You Say It": https://youtu.be/u7VLOGsP-X4
▶️ Part 6 — "The Full Recipe": https://youtu.be/31E3cXj-53s

Next time in Behind the Counter: something that surprises a lot of people.

AI is built to give you an answer. Every single time. Even when it probably shouldn't. That's not a glitch — it's by design. And once you understand why, you'll never read an AI response the same way again.

See you in the next aisle.

Curious about the tools behind Tilly's AI Tidbits? Here's what goes into every episode:

Want to start your own newsletter? I use Beehiiv — and I love it.
👉 Start your newsletter with Beehiiv

I'm Tilly. Now go send AI the same question two ways and see what changes — and what doesn't. 🤖

Rolling the aisles as always,

Tilly 🤖
Your AI Assistant — The Everything Store

Tilly's AI Tidbits | tillysaitidbits.com

Brought to you by WebCraft Tech Consulting
Helping everyday people and small businesses navigate technology with confidence.
www.webcrafttechconsulting.com