Meta AI, A to Z

Where Meta AI lives

Rosa has run Crumb & Craft, a neighborhood bakery, for nine years. Her marketing plan is a corkboard by the register and whatever her teenage niece posts on the shop's Instagram between classes. When her niece Maya came home for the weekend, Rosa asked the question she'd been embarrassed to ask: "Everyone keeps saying AI. Where even is it?"

Maya held up her phone, opened Instagram, and tapped the little blue circle at the top of her direct messages. "That's Meta AI," she said. "It's in here. It's also in Facebook, in Messenger, and in WhatsApp. You don't download anything. You already have it."

Rosa was skeptical in the way busy people are skeptical of anything that sounds like a project. That evening, after the dinner rush, she opened Facebook on her own phone, tapped the search bar, and saw the Meta AI prompt waiting there. She typed: "Write a friendly Facebook caption for a photo of fresh sourdough loaves, warm tone, under 40 words, end with a question." Ten seconds later she had something better than the "Fresh bread today!" she'd been posting for three years.

What she figured out over the next week is the habit that makes Meta AI different from every other AI tool: ask in the app where you'll use the answer. Need a caption for Instagram? Ask inside Instagram, then post it. Need a reply for a WhatsApp customer? Ask inside WhatsApp, copy it, send it. The answer arrives exactly where the work happens.

The following week, Rosa discovered the real magic wasn't the first answer, it was the follow-up. When a caption came back too long, she just typed "shorter." When it sounded stiff, she typed "more playful, like I'm talking to a neighbor." No new prompt, no starting over, just talking, the way she'd talk to Maya. Within two weeks she was having full conversations with it: draft, tweak, tweak again, post. Her niece's verdict, delivered with appropriate teenage gravity: "See? You're a prompt engineer now."

The takeaway: Meta AI lives where your customers already talk to you, Facebook, Instagram, Messenger, WhatsApp. No new account, no new app, no learning curve for the tool itself. The only learning curve is learning to ask well, which is what the next seventeen lessons are for.

No new app needed

Marcus is 54, a master plumber, and the proud owner of a phone full of apps he opened exactly once. His dispatcher, Tanya, has been trying to get him to "use AI" for months. His answer is always the same: "I'm not downloading another thing."

So Tanya tried a different pitch. "You don't download anything," she said. "It's in WhatsApp, the green one you already use for job photos." She took his phone, opened WhatsApp, and tapped the Meta AI tab at the bottom. "Ask it for the thing you type twenty times a week."

Marcus typed: "Write a short text telling a customer I'm running 30 minutes late, professional but friendly, from Marcus at Marcus and Sons Plumbing." He read the result, changed one word, "apologies" felt stiff, "sorry about that" felt like him, and saved it in his phone's notes.

Over the next week he built a little collection: the running-late text, the "job's finished, everything's working" text, the "here's your invoice" text, the Saturday-hours reply. All drafted inside WhatsApp, all saved where he could grab them in two taps. Tanya stopped asking him to download things.

Tanya, meanwhile, started using it for her half of the business. She drafts the appointment confirmations now, "Hi, this is Marcus and Sons confirming tomorrow between 9 and 11", and Marcus just reads and sends. When a draft comes back too formal, Marcus has learned the two magic words: "more casual." When it's too chatty: "more professional." He's not learning AI, he'd tell you. He's learning to give directions, which, as a man who runs a crew of four plumbers, turns out to be something he already knew how to do.

The takeaway: the best tool is the one you'll actually open. Meta AI rides inside apps already on your home screen, so there's nothing to install and nothing to learn except asking. (If you ever want a bigger keyboard, there's also a standalone Meta AI app and meta.ai on the web, same assistant, no extra account, but you don't need either to start.)

First questions

Priya owns Lumen Hair Studio, and her first two questions to Meta AI taught her everything about how to talk to it. Question one: "Give me 5 Instagram caption ideas for balayage transformation photos, playful tone." The answers were excellent, specific, usable, on-brand. She posted one that afternoon.

Question two: "What's the best hair color brand?" The answer was generic, hedged, and useless, the kind of thing you could find on any forum. Priya stared at the two answers side by side and saw the difference immediately. The first question had given the AI her context: her service, her platform, her tone. The second had given it nothing, so it gave nothing back.

She developed a habit she's used ever since: every prompt gets three ingredients. What the thing is. Who it's for. How it should sound. Her captions went from "give me a caption" to "Write a caption for a bridal updo photo, aimed at brides planning 2026 weddings in Columbus, elegant tone, end with a booking call to action." The difference in output quality was embarrassing, in her favor.

The three-ingredient habit got its real test with a hiring post. Priya's first try, "write a hiring post", produced something that could've been for any business on earth: "Join our dynamic team!" Her second try: "Write a Facebook hiring post for a front-desk receptionist at a busy hair salon in Westerville, aimed at friendly, organized people who love the beauty industry, warm and upbeat tone, mention Saturdays required." The second one sounded like her shop, it even mentioned the Saturday rush by name. Same tool, same ten seconds. The only difference was what she put in, and three qualified applicants replied within two days. She now runs every important post through the three-ingredient check before she even opens the AI, what, who, how, and calls it the closest thing to a cheat code this course has.

The takeaway: Meta AI is a mirror. Vague in, vague out. Give it the what, the who, and the how, and it gives you something you can actually post. This one habit improves every lesson that follows.

Post ideas

Dan owns Arctic Air Pros, an HVAC company, and his Facebook page told the story: two posts in all of 2024, both reading "Call us for service!" He knew he should post more. Every time he opened the blank box, his mind went blank with it.

On the first Sunday of June, coffee in hand, he opened Meta AI in Messenger and typed: "Give me 10 Facebook post ideas for an HVAC company in Columbus for June. Mix: money-saving tips, behind-the-scenes, a summer tune-up offer, and customer stories. One line each."

What came back felt like someone had been watching his business for a month: "5 signs your AC is working too hard," "What our techs actually find in your filters (photo)," "June tune-up special, $79," "The Johnsons' 22-year-old unit that made it one more summer." He picked eight, asked for captions for his top three, and scheduled the whole month in forty minutes.

By August it was a ritual: first Sunday, coffee, forty minutes, month handled. His page went from a ghost town to the place neighbors tagged each other under. Two of those tags turned into service calls.

Dan added one more move that doubled the value of his Sunday sessions. After getting his ten ideas, he asks: "Which 3 of these 10 will get the most engagement from homeowners over 40 in Columbus, and why?" Meta AI ranks them and explains its reasoning, the money-saving tip wins because it promises a concrete number, the filter photo wins because it's visual and slightly gross, in the best way. He leads the month with those three and treats the rest as filler. The batching gives him quantity; the ranking gives him a starting order. Forty minutes, one coffee, zero blank boxes. His only regret, he'll tell you, is the two years he spent posting nothing while waiting to feel "ready", the system took less time than a single service call.

The takeaway: consistency beats brilliance. One batching session a month, inside an app you already use, ends the blank-page problem forever. Ask for the ideas in bulk, then go deeper only on the winners.

Caption writing

Alicia's coffee shop, Daybreak Coffee, takes gorgeous photos. Golden-hour latte art, the morning rush through the window, her barista's sleeve of tattoos pouring a rosetta. And every caption read the same: "New seasonal latte! Come try it!"

A friend who does social media for a living showed her the trick: brief Meta AI the way you'd brief a copywriter. Don't just say "write a caption." Describe the photo. Say who it's for. Name the vibe. Set a length. Demand a call to action.

Alicia's next prompt: "Write an Instagram caption for a photo of a honey-lavender latte with heart latte art on a rustic wooden table, morning light. Audience: young professionals in Westerville. Cozy, upbeat tone. Under 50 words. End with a question."

Before: "New seasonal latte! Come try it!" After: "Honey, lavender, and a little morning magic. Our spring latte is here through May, which table are you claiming this weekend?" Same photo. Different business. The second one got eleven comments; the old style got one, from her mom.

Alicia turned it into a game she plays every Friday. Same photo, three captions: one funny, one elegant, one urgent. She reads all three out loud to her barista, they pick the winner together, and that one goes up. Some weeks the funny one wins; some weeks elegant takes it. The point was never which style wins, it's that she now has three good options in thirty seconds instead of one tired line she's posted forty times. Her comment rate tripled in two months, and her staff started keeping score over whose pick won. The Friday caption vote is now a Daybreak Coffee tradition. New baristas learn the game on their first Friday, which means Alicia hasn't written a caption alone in months, and the page has never sounded more like the shop itself.

The takeaway: the photo earns the scroll-stop; the caption earns the comment. Give the AI the details you'd give a human writer, the photo, the audience, the tone, the length, the ending, and it writes like one.

Hashtag help

Nadia opened Form & Function, a women's strength-training studio, and did what everyone does: #fitness #gym #love #instagood. The result was a parade of bot comments, "Great post! Check my page!", and zero local clients.

She asked Meta AI in Instagram DMs: "Give me 15 hashtags for a women's strength-training studio in Westerville posting about beginner classes. Mix broad, niche, and local tags." Then she asked the follow-up that mattered: "Explain why each group works."

The logic clicked. Broad tags like #fitness are crowded oceans where her post drowns in seconds. Niche tags like #beginnerstrengthtraining are ponds where she can actually be seen. Local tags like #westervilleohio and #columbusfitness put her in front of people who can physically walk through her door, the only audience that matters.

She also learned to check before she commits: tap any hashtag and look at its post count. Millions of posts means skip it. Thousands to low hundreds of thousands is the sweet spot. And eight to twelve well-chosen tags beat thirty every time, Instagram reads hashtag-stuffing as spam.

Nadia refreshes the set every quarter, because hashtags have seasons. In January she asks: "Update my hashtag set for new-year beginners, people in Westerville searching for a gym in January." Out go the summer tags, in come resolution-season terms and local new-year variants. It takes four minutes. Her January beginner-class posts started reaching people who were actively, specifically looking for exactly what she sells, in the month they look hardest. She also drops any tag that stops performing, if a hashtag brings bots two weeks running, it's out, no sentimentality. Her rule of thumb now: if she can't explain in one sentence why a hashtag is on the post, it doesn't belong there. The bots are gone, by the way, spammers don't bother with tags like #westervillewomenwholift.

The takeaway: hashtags are a filing system, not decoration. File yourself where your actual customers browse: a focused mix of broad, niche, and local, checked for size, kept under a dozen.

Quick replies

Greg runs Honest Auto, and he types the same sentences so often his thumbs have memorized them: "Brake pads for most sedans run $180 to $260 including labor, want me to check your specific model?" "Yes, we're open Saturdays 8 to 2." "The diagnostic is $95, and it's applied to the repair if you go ahead." Twenty times a week, easy.

One slow Tuesday he opened Meta AI in Messenger and asked: "Write 6 short, friendly replies for an auto repair shop: brake price question, Saturday hours, diagnostic fee explanation, oil change wait time, 'can I wait for it,' and thank-you after service. Two sentences max each, plain words."

He tweaked the wording to sound like himself, "swing by" instead of "visit our location", and saved them as text replacements on his phone. Now typing "brkpd" expands into the full brake-pad answer. Typing "sathrs" gives the Saturday hours. A thirty-second question gets a ten-second answer, and it reads like he wrote it fresh every time.

The templates needed one lesson Greg didn't expect. A customer replied to the brake-pad text: "That's more than the dealer quoted me." Greg's first instinct was to get defensive. Instead he asked Meta AI: "Write a calm reply to a customer who says our brake price is higher than the dealer. Explain we use OEM parts and include a 2-year warranty, no defensiveness, invite them to call." The customer called. Then he booked. Greg's rule now: templates handle the routine, the moment a conversation turns delicate, he drafts fresh, reads it twice, and sends it himself. He's since built a second set for estimates, "here's what we found," "here's what it costs," "here's what can wait", because the same principle works anywhere he repeats himself. His service writer uses them too, so the whole shop now sounds like Greg on his best day, even at 5pm on a Friday.

The takeaway: every question you've answered fifty times is a template waiting to happen. Draft the set once with Meta AI, save them where your thumbs live, and routine questions stop eating your day.

WhatsApp Business

Beatriz's flower shop, Petal & Stem, takes most of its orders over WhatsApp. Between "Do you deliver to Dublin?" and "Can I get peonies by Friday?" she was answering the same questions all day, often at 10pm, bleary-eyed, thumbs aching.

The fix started with switching to the free WhatsApp Business app, separate from regular WhatsApp, works with the same number, which gave her three superpowers: a greeting message that auto-sends to first-time messengers, an away message for evenings and Sundays, and quick replies for her most-typed answers.

She drafted all of it with Meta AI, right inside WhatsApp: "Write a warm greeting message for a flower shop's WhatsApp: thank them for messaging, say we reply within a few hours, ask what occasion they're shopping for. Two sentences." Then an away message: "Thanks for reaching out, we're closed for the evening and will reply first thing in the morning. For anything urgent, call [NUMBER]." Then ten quick replies, from delivery areas to funeral-arrangement timelines.

One honest note she wants every business owner to hear: Meta AI drafts the words, Beatriz still taps send. It doesn't auto-reply to her customers, and she likes it that way. She's the voice; it's the ghostwriter. Two weeks later, routine questions get answered in seconds, the 10pm messages get a kind away message instead of her exhausted thumbs, and not one customer noticed a change except that she "got faster."

The away message earned its keep on a Sunday in May. A bride, panicking about Saturday delivery for a wedding, messaged at 9pm while Beatriz was at her sister's birthday dinner. The away message replied instantly: thanks for reaching out, we're closed this evening, for urgent wedding orders call this number. The bride called the number, Beatriz's cell, listed for exactly this case, they sorted it in five minutes, and Monday morning the order was confirmed. The system didn't replace Beatriz. It just made sure a panicking bride reached a human instead of silence, and that bride has ordered three more times since.

The takeaway: WhatsApp Business plus drafted templates turns your busiest channel into your calmest one. Set the greeting, the away message, and the quick replies once, then stay human on the parts that matter.

Comment responses

Tom posts satisfying before-and-after lawn photos for Greenline Landscaping, and the comments roll in: "Nice work!" "Looks great!" He means to reply. He doesn't. For weeks, the comments sit there like unanswered mail.

Then this one: "Looks nice but $400 seems steep for mowing." It sits for six days. By the time Tom sees it, eleven people have liked it, and the silence underneath it reads as agreement.

That stung enough to build a system. With Meta AI's help, he made three buckets. Thank: "Thanks, Denise! That slope was a beast, glad you like how it turned out." Answer: genuine questions get genuine answers, fast, because a question in the comments is a sales call you didn't have to make. Defuse: price complaints get calm, factual, kind, "Fair question, that covered the mow, edging, and the full leaf cleanup. Happy to walk you through it anytime: (614) 555-0134."

He asked Meta AI: "Write 4 warm replies to positive comments on a landscaper's Facebook page, each different, one or two sentences." And separately: "Write a calm, professional reply to a comment saying our price seems high. No defensiveness, invite a phone call." He keeps both lists in his notes app.

The defuse reply paid for the whole system. The man behind the "$400 seems steep" comment called the number Tom had left. Tom walked him through the leaf cleanup, the edging, the disposal, all the parts of the job invisible in a photo. The man booked. Then he became a regular, every other week, $400 a visit. Tom did the math once, late at night: that single calm reply, drafted in two minutes with Meta AI's help, was worth roughly ten thousand dollars a year. He has never left a comment unanswered since, and his reply rate is now something he brags about.

The takeaway: every comment is a public conversation, and silence is an answer too. Replying tells the algorithm your page is alive, and tells customers you're listening even when the comment is uncomfortable.

Describing what you want

Ken owns Sudsbury Grooming, and his first attempt at AI imagery was not glorious. He opened a Messenger chat with Meta AI and typed: "imagine a dog getting groomed." What came back was a generic cartoon dog in a generic salon, technically a dog, technically grooming, completely useless for his page.

A groomer friend showed him the formula on a napkin: subject, details, setting, light, style, and what to avoid. Ken tried again: "imagine a fluffy cream golden doodle getting a bath in a bright modern grooming salon, blue tile walls, morning sunlight through big windows, groomer in navy apron gently holding the dog, photorealistic, cheerful, no text, no watermark."

Night and day. The second image looked like his shop on its best morning. He posted it with "Bath day at Sudsbury" and booked two appointments from the comments, people recognized the blue tile.

Two mechanics worth knowing: the word "imagine" at the start of your message is the trigger, that's how you ask Meta AI to generate an image instead of answering in words. And it works in Messenger, Instagram DMs, and WhatsApp alike, so you can create the image in the app where you'll post it.

Ken's second discovery was iteration. The first good image was close, but the lighting felt flat, so instead of starting over, he just typed in the same chat: "same image but warm sunset lighting through the windows." Meta AI kept the dog, the salon, the composition, and changed the light. Then: "same but the groomer smiling at the camera." Three tweaks, same thread, each building on the last. He'd been treating it like a vending machine, one prompt, one picture. It's more like an art director who never gets tired of revisions, and the thread remembers everything you've already decided.

The takeaway: image generation is a camera that only takes direction. Vague direction, vague photo. Art-direct it like a photoshoot, subject, details, setting, light, style, exclusions, and it will surprise you.

Business uses

Lena manages the front desk at Riverside Dental, which wanted to promote a free whitening consultation. Their designer quoted $150 and a week. Lena needed it by Friday.

She opened Meta AI and generated a clean, bright background, soft blue gradient, subtle sparkle, no people, no text, then added the words herself in Canva: "Free Whitening Consult, This Month Only." Total cost: zero. Total time: twenty minutes. The post outperformed their last three designer-made graphics.

That became the rhythm. Seasonal backgrounds she generates herself: pastel flowers for spring, pumpkins for October, snowflakes for December. "Did you know" illustration posts about brushing and flossing. Real photos of the actual office, the actual team, the actual waiting room, those stay real, always.

Her rule came from watching a colleague at another practice learn it the hard way: generate art, photograph reality. A generated illustration of a sparkling tooth? Perfectly fine. A generated "before and after" of a patient's smile? Never. That's someone's health, and fake clinical imagery doesn't just look wrong, it destroys trust the moment anyone looks closely.

Lena's smartest move was asking for consistency. A feed full of clashing graphics looks amateur, so she asked: "Generate 4 seasonal backgrounds for a dental office, spring, summer, fall, winter, all soft blue gradients with subtle seasonal details, same minimal style, no people, no text." Four images, one visual family. Her feed suddenly looked designed instead of assembled. When she needs a new one, she references the set: "same style as the spring background, but with pumpkins." The AI remembers the thread, the brand stays coherent, and patients have actually commented that the page "looks expensive." The front desk now generates the monthly special graphic in-house, every single month, and the $150-a-pop designer is reserved for the big rebrand they actually need her for.

The takeaway: Meta AI's image generation is a free junior designer for backgrounds, seasonal art, and illustrations. Pair generated art with your real photos, keep clinical and documentary imagery strictly real, and nobody can tell where the design budget went.

Limits

Sam is a realtor, and staging a new listing gave him what felt like a clever idea: generate a gorgeous kitchen photo for the flyer. He typed "imagine a luxury kitchen, marble island, pendant lights, photorealistic." What came back looked stunning at thumbnail size.

Then he zoomed in. Seven cabinet handles on a single drawer. A faucet emerging from the wall at a structurally impossible angle. And the word "LUXURY" misspelled in the tile backsplash. He laughed, then felt a chill, he'd nearly put that on a flyer with his name and license number on it.

He learned three hard limits that afternoon, and they apply to every business. First: text inside generated images comes out garbled more often than not, if you need words, generate the art and add the words yourself afterward. Second: hands, faces, and fine details glitch; always zoom in before you post. Third, and this one matters legally: never present a generated image as a real photo of a real place. A fake listing photo isn't marketing. It's a liability with your signature on it.

Sam still uses image generation constantly, neighborhood lifestyle illustrations for market updates ("fall in Clintonville"), dream-home vision boards for buyers, seasonal greeting cards. He just knows which lane it's in now.

Sam found the lane where generated images genuinely shine: imagination, not documentation. A buyer loved a house but couldn't see past the dated living room, brown carpet, heavy drapes, 1994. Sam generated an illustration: "same living room layout reimagined in modern farmhouse style, light oak floors, white walls, linen sofa, bright and airy illustration." He labeled it clearly on the slide: "Vision concept, not the actual room." The buyer made an offer that week. The art helped them dream; the real photos closed the deal. That's the division of labor, and Sam now uses it for every stale listing.

The takeaway: generated images are art, not evidence. Use them to decorate and illustrate. Never use them to document, and always inspect closely before anything goes public.

Script outlines

Elena runs Sparkle Home Co., a cleaning service, and her first Reel was a 40-minute deep-clean of a trashed rental, narrated live, unedited. She posted it proudly. Two hundred views, and she suspects half were family.

A friend who does social media for a living gave her the uncomfortable truth: nobody watches forty minutes. People watch thirty seconds, if the first three seconds earn it. "You don't need a better phone," her friend said. "You need a plan."

Elena opened Meta AI in her Instagram DMs: "Outline a 30-second Reel for a house cleaning business: satisfying oven-clean transformation. Hook for the first 3 seconds, 3 quick beats, call to action at the end. Energetic, no jargon." What came back was a shooting script: Hook, "This oven hasn't been cleaned in 4 YEARS." Beat one, the horrifying before shot. Beat two, the product, one wipe, the sizzle. Beat three, the reveal. CTA, "Follow for the bathroom the landlord said was hopeless."

She refilmed the same oven with the outline taped to the cabinet. Twelve thousand views. Three quote requests in the comments. Same oven, same phone, same Elena, the only thing that changed was the plan.

Elena's second video applied the same outline to a grout transformation, and she added one new trick: on-screen text. She asked Meta AI: "Suggest 4 short on-screen text overlays for a 30-second grout-cleaning Reel, one per beat, under 6 words each, punchy." She got "THIS GROUT IS 10 YEARS OLD," "ONE PRODUCT. ONE BRUSH.", "WAIT FOR IT…", "BOOK YOUR DEEP CLEAN." She'd learned that most people watch without sound, so the text carries the story when the audio can't. Views doubled again. The outline got them watching; the text kept them watching; the call to action got them booking. Her booking calendar has a two-week wait now, and every new video starts the same way: outline taped to the cabinet, phone propped up, thirty seconds.

The takeaway: the camera was never the problem. Outline first, film second, every time: a hook for the opening seconds, three tight beats, and a call to action that gives viewers a reason to follow.

Trends explained

Hannah owns Willow & Wren, a women's boutique. She kept seeing the same dramatic audio clip under dozens of videos, swelling music, people revealing things with a flourish. She watched five of them and understood exactly nothing about what she'd do with it. She's 41, sells cardigans, and the whole thing made her feel approximately one hundred years old.

Instead of guessing, she asked Meta AI to translate: "Explain the current dramatic-reveal audio trend on Instagram Reels like I'm a smart adult who doesn't live on the app. Then give me 3 ways a women's clothing boutique could use it this week."

The explanation gave her the key insight she's used ever since: a trend is a format, not a topic. The format here was "dramatic reveal", and her versions wrote themselves. Reveal the fall collection piece by piece. Before-and-after of a styled outfit. What $200 buys at her store versus the mall, revealed with the same flourish.

She filmed one that afternoon. It wasn't viral. It got forty comments from actual customers, women who shop with her, tagging friends who might too. Which, she decided, is better than viral.

Hannah's second translation was a "get ready with me" video, the format where creators get dressed on camera while chatting. Her version: "Get dressed with me for date night, three outfits from our new arrivals, under $150 each." She filmed it in the stockroom between customers, phone propped on a shoebox, talking the way she talks to regulars. Twenty minutes, one take. Two of the three outfits sold out that weekend. The trend supplied the container; the shop supplied the content. She now keeps a running list of trends on her phone and asks Meta AI to translate the newest one every Monday morning.

The takeaway: you don't need to be trendy; you need to borrow the format and keep your message. Whenever a trend confuses you, ask Meta AI to explain it in plain words and translate it into three ideas for your specific business.

Hooks that work

Chris runs Shield Pest, and every one of his videos started the same way: his logo, fading in, for four seconds. His watch-time graph fell off a cliff at second two. Every single time. He was essentially paying the first four seconds of every video to show people a logo they'd already seen on his truck.

He asked Meta AI: "Give me 10 opening lines for short videos about termites, each under 10 words, designed to stop the scroll. Mix: surprising result, common mistake, and direct question." Back came lines like "There's a 1-in-5 chance these are in your walls." "Stop stacking firewood against your house." "Hear clicking in your walls at night?"

Over the next month he tested three formulas. Result first: open on the damage, then explain how it happened. Mistake: "The number one thing homeowners do that invites ants." Question: "Why do wasps love your deck every August?" His average watch time tripled, not because the videos got better, but because the first three seconds finally earned the remaining twenty-seven.

Chris made it scientific. Same tip, "keep mulch away from your foundation", filmed three ways across three weeks, each opening with a different hook: the result ("Termites caused $5 billion in damage last year"), the mistake ("You're inviting termites with your landscaping"), the question ("What's eating your house right now?"). He compared views at the three-second mark, not total views, that's the only number the hook controls. The mistake-hook won by a mile. He keeps the winners in a notes file titled, with total seriousness, "Hooks That Slap," and he tests every new video idea against it before filming. His latest test: the same winning hook with and without captions burned in, captions won, because most of his audience watches with the sound off. The notes file grows every month, and new hires study it like a playbook.

The takeaway: a hook isn't clickbait; it's a promise. State the payoff, the mistake, or the question in under ten words and three seconds, then spend the rest of the video keeping the promise.

Fact-checking

Sofia owns Glow Studio, a med spa, and she was rushing out the door when Meta AI handed her a lovely draft: "Summer glow-up special! Book this week, we're open Sundays now!" She copied it, posted it, and left.

Sunday morning, a client rang the bell. Glow Studio is closed on Sundays. It has always been closed on Sundays. Meta AI had invented "open Sundays now" out of thin air, confidently, in perfect grammar, as if it had checked.

Sofia apologized in person, honored a discount she hadn't planned to give, and made a new rule that afternoon: Meta AI writes the draft; Sofia checks every fact. Hours, prices, dates, addresses, any claim about results, all of it, every time, before anything goes public. She even made Meta AI help with the checking: "Here's my draft post. List every factual claim in it, hours, prices, dates, promises, as a checklist I should verify before posting."

She understands now why it happened. Meta AI doesn't know her business; it knows language patterns. When it hits a gap in its knowledge, it fills the gap with something plausible-sounding rather than admitting the gap. That's fine for brainstorming captions. It's dangerous for anything a customer might act on.

The checklist earned its keep the very next week. Meta AI drafted a post about the "Spring Refresh package, $99!" Sofia ran her checks: hours right, price… wait. The package was $129. The AI had invented a $99 price out of the same thin air as the Sunday hours. She fixed it before posting. That was the moment the rule changed from "good idea" to "non-negotiable", two catches in two weeks, both in the exact categories the checklist covered. She now tells every new hire: the AI writes the sizzle, and we check the steak. Nobody posts without the checklist.

The takeaway: AI is the intern with perfect grammar and no memory of your business. You're the editor-in-chief. Nothing ships without your eyes on the facts, hours, prices, dates, claims, all of it.

Privacy sense

Paul is an insurance agent with 300 clients and a genuinely nice idea: personal birthday messages on WhatsApp, one per client, every morning. At 11pm one night, an even more efficient idea arrived: paste the whole client list into Meta AI and have it write all 300 at once.

He stopped himself, barely, and thought about what he'd actually be uploading: names, phone numbers, birth dates, policy types. His clients' private lives, typed into a chat window, to save an hour of copy-paste.

His new rule is simple enough to fit on a sticky note: the public test. If the content is something he'd happily post on his business Facebook page, marketing ideas, captions, generic advice, a template with a [NAME] placeholder, Meta AI is fine. If it's something he'd never post publicly, client names, phone numbers, financial details, health information, it never goes in the chat box. Not in WhatsApp, not in Messenger, not anywhere.

What he did instead worked better anyway: "Write 5 warm birthday message templates for insurance clients, professional but personal, each different, with [NAME] as the only placeholder." He pastes the templates into WhatsApp and fills in the names himself, in the app where the messages were going regardless.

Paul's actual workflow turned out better than the shortcut. Every morning he opens his five templates, pulls up the day's birthdays, usually eight to twelve, and personalizes each one in about a minute: the name, plus one true line about their policy ("Hope the new truck's treating you well, Mike!"). Fifteen minutes total. The replies are the best part: "How did you remember?!" every single time. Nobody suspects a template was involved, because the personal line is real and specific. He saved the hour without spending anyone's privacy to do it, and his retention numbers noticed.

The takeaway: Meta AI is a public square with a very helpful stranger in it. Wonderful for public-facing work. Never bring private business there, if you wouldn't post it, don't paste it.

When to use a bigger tool

Rachel runs Ledger & Co., a bookkeeping practice, and she loves Meta AI for quick work, captions, replies, ideas, the fast social stuff. So when a client asked for a year-end tax checklist, she tried to build the entire twelve-page document in a WhatsApp chat with Meta AI.

It went badly. The chat lost track around page four. The formatting was a mess of run-on paragraphs. She couldn't get a clean, saveable document out of it no matter how she asked. She spent longer fighting the chat window than the checklist would have taken to write from scratch.

That failure taught her the division of labor she now explains to every business owner who asks: Meta AI, inside your social apps, is for fast, social, short work, captions, replies, ideas, hashtags, quick drafts. Seconds, on your phone, right where you'll use the answer. A bigger tool, ChatGPT, Claude, or Gemini on the web, is for long documents, spreadsheets, analysis, anything you need to save, edit, structure, and format properly.

Neither is "better." A bicycle isn't worse than a truck; it's just the wrong vehicle for moving house. Her rule now: if the answer fits in a post, a message, or a comment, Meta AI on her phone. If it needs pages, structure, or math, the bigger tool on her laptop.

Rachel's final workflow is the hybrid this whole course has been building toward. When that year-end checklist came up again, she did the thinking on her phone in Meta AI, "Give me 10 sections for a small-business year-end tax checklist", then built the actual twelve-page document in the bigger tool on her laptop, where she could structure it, format it, and save it properly. Phone for thinking, laptop for building. Two tools, two lanes, zero fights with a chat window. She finished in an afternoon, billed it the same day, and the client called it the clearest checklist they'd ever received.

The takeaway: the real skill isn't learning one AI, it's learning which AI for which job. Meta AI owns the quick social lane completely. Let it own that lane, and hand everything else to the tool built for it.