Zuhair Lakhani - How to Make Viral Content 100% Using AI in 2025

Borrowed the Supreme / sneaker-bot drop playbook for his dad's new restaurant: a reservation waitlist, with reservations dropped every Wednesday at 4:00 p.m. and announced in a local Facebook group. On the first drop only ~20 people signed up for 100 spots, but he declared it full anyway: better to look full and have people chase you than to look available. The waitlist also let them control table sizes (e.g. seat a party of 8 at a 7-seat table plus a chair) to fill capacity. TikToks, Google Ads and Facebook also ran.

Posted in a hyper-specific local Facebook group (e.g. a Dallas Restaurant Group) and encouraged diners to write reviews there, since local groups target the right people precisely (like a niche subreddit for a consumer app). Each 'I just tried this, the food was amazing' post created FOMO for other members.

Took an unused poster design made for a client, put it on a poster mockup (posters were easy to source fast, which matters with TikTok Shop's two-day delivery), listed it on TikTok Shop and made one video with AI (ElevenLabs voice + Midjourney images). When it got 3 sales, he left a $100 deposit with a local print shop so they would print on each call and bill his card afterwards.

Ran sub-accounts (not the main account) purely for views and traffic. Format: 'cool stuff you need in your room' slideshows listing several products, one of which is always your own poster. Another account posted ~8 pictures of a Porsche 911 with the poster as the last picture, so viewers ask where to get it. Images pulled manually from Cosmos/Pinterest, assembled with Figma components by a VA. Prove the format works manually before automating.

Created fully synthetic creators: contract people's likeness for an image set, prompt until one type of person comes out consistently, then train a LoRA on that person (Flux, Pixelfed). Each account posts a full slideshow 1-2 times a day, with the person telling a problem -> solution story that works the product in naturally. Ran 25 such accounts. AI does ~95% of the work; a human does the final ~5% check before anything posts.

Slide 1 is a hook naming the pain ('I used to struggle with tracking my calories'), fronted by an AI creator, a real creator or a faceless slide (only slide 1 is AI-generated). Middle slides give real but high-effort solutions (e.g. weighing meals on a scale); a later slide (e.g. #5) is your app as the cheaper, lower-effort fix. Each slide position has its own image bank: if slide 2 always clarifies the problem, a 50-image bank can run ~300 slideshows.

Having been one of 25 hired college creators (30 videos in 30 days for $500-1,000/month, one full-time manager, weekly calls, little accountability), he judged the ROI poor and instead produced the same kind of content with AI for his agency's brands. Posting was the bottleneck: he used Android emulators (BlueStacks on Windows) from 2023, which TikTok's device fingerprinting now catches.

Each account lives on a real phone controlled by custom software that swipes, reposts and comments like a person (emulators get caught by device fingerprinting). Account setup: describe the product and target (e.g. 20-44, mostly male), AI generates usernames, names, bios and gender, and each account gets a different set of niche search terms (self-care, self-improvement, lifestyle, motivation) so accounts aren't identical but don't vary too much. Warm-up searches those terms, screenshots what it watches, asks an LLM whether it's relevant, then reposts/comments if so or keeps swiping. Specific warming algorithms are not shared.

Find formats by searching your product's terms on TikTok (e.g. StudyTok for a study tool) and picking viral hits from the last 1-2 months. Paste the TikTok link to generate a template, write the template prompt by hand (e.g. 'five Spanish phrases you should know'), then generate variants: AI writes each slide's text, images are randomized from an image bank, hooks and CTA ('Download this app') can be changed in bulk. Always change something versus the original so you have a chance to win. For hook + demo videos, auto-match a library of hook clips with app screen recordings and add copy rather than paying creators. Post results feed back into what gets made next.