AI & foodmenu strategyfood trends

AI-Generated Menus: Using Vertical-Video Data to Discover What Diners Want

fflavours

2026-02-03

10 min read

Use vertical-video signals and AI to prototype limited-time menu hits—fast. Learn a practical 10-step playbook for 2026.

Hook: Stop guessing — let what people watch tell you what they’ll eat

Restaurants spend months — sometimes seasons — crafting menu ideas that might never find traction. The pain is familiar: you need fast, low-risk ways to prototype dishes, test real diner appetite, and move quickly when something sticks. In 2026, the clearest answer is sitting on phones: short-form, vertical-video consumption data. When combined with platforms like Holywater for data-driven IP discovery and Google’s Gemini guided learning for rapid iteration, restaurants can prototype, test-market, and scale limited-time offers (LTOs) faster than ever.

The evolution in 2026: why vertical video data matters now

Short-form vertical video dominated the last half-decade of content discovery. In early 2026, Holywater — backed by Fox — raised fresh capital to expand an AI-first vertical platform that does more than stream: it surfaces what formats, flavors, and moments are resonating with mobile audiences (Forbes, Jan 2026). At the same time, generative AI systems like Gemini matured from research assistants into guided learning and operational partners that help teams turn signals into repeatable processes (Android Authority, 2025–26).

Put simply: audiences now reveal preferences in the act of watching and rewatching vertical shorts. That raw behavioral data is gold for menu teams who want to test ideas in market without heavy upfront investment.

How vertical-video consumption predicts dining choices

Video consumption is an active signal: watch time, rewatches, comments, duets, and user-generated remixes show emotional engagement — the same force that drives a diner through your door. Key behaviors to monitor:

Completion rate: People who watch to the end most often try to replicate or seek out the item.
Rewatch and slow-motion replays: Indicates sensory appeal — crisp crunch, melty cheese, steam.
Remixes and UGC: When viewers post their own version, adoption potential jumps.
Comment sentiment: Requests for recipe, location, or price indicate intent to purchase.
Local clustering: Concentrated engagement in a geography signals a nearby market fit.

These signals map to real-world behaviors: search interest, KR (keyword request) spikes, and local foot traffic. That mapping is what smart restaurants can exploit to run fast, low-cost experiments with LTOs.

Holywater’s data-driven IP discovery — and what restaurants can borrow

Holywater’s recent expansion (Forbes, Jan 2026) is built on two ideas: 1) short-form, serialized content creates reproducible formats and 2) AI can discover which formats are IP-worthy by analyzing consumption patterns. Restaurants don’t need the full Holywater stack to use the same logic. The principle is transferable:

Identify repeatable formats: Is the audience gravitating to recipe reveal, ASMR crunch, heritage storytelling, or micro-drama about ingredients? Formats predict how a dish will travel in culture.
Cluster flavors and motifs: AI can group trending terms (e.g., yuzu-kimchi, hot honey wings, charcoal brioche) from thousands of short videos to surface emergent flavor clusters.
Time-to-signal: On vertical platforms, trends accelerate: what shows strong engagement for two weeks is likely to sustain interest if paired with the right distribution.

Actionable takeaway: Build a simple vertical-video feed monitor that captures trending formats + flavor clusters in your trade area and rank them by intensity, velocity, and local concentration.

Gemini as your creative lab partner

Gemini and similar guided-learning LLMs now do more than write copy: they synthesize behavioral signals, produce recipe iterations, cost ingredients, scale for a kitchen line, and generate creative briefs for short-video shoots. Android Authority’s tests in 2025 showed these models excel at curating guided learning paths — a capability restaurants can use to compress the R&D cycle.

Here’s what Gemini-like systems can handle for your menu experiment:

Concept synthesis: Turn trend clusters into 3–5 menu concepts prioritized by ease of execution and margin potential.
Recipe scaling: Convert a 1-plate recipe into batch quantities with yield-adjusted costs and prep steps.
Training scripts: Produce quick video-friendly SOPs for cooks and front-of-house.
Creative briefs: Generate vertical-video shot lists, hooks, and sound suggestions optimized for watch-time.

Follow this 10-step process to use vertical-video data and AI to prototype LTOs in 2–6 weeks.

Listen for trends (days 0–3): Pull 2–4 weeks of vertical-video data from platform APIs, Holywater feeds (if accessible), and public TikTok/Shorts/Reels trends. Look for flavor clusters, format types (recipe, ASMR, story), and geography. Score by engagement velocity and local intensity.
Filter by operational fit (day 3): Remove concepts that require unfamiliar equipment or rare ingredients. Prioritize ideas that can be made with existing line tools in under 6 minutes per plate.
AI concept generation (days 3–4): Use Gemini to generate 5 concept variants per flavor cluster with estimated food cost, suggested garnish, and a 30-second vertical-video creative brief.
Rapid prototyping (days 4–7): Cook scaled 4–6 plate trials. Use staff tasting and a small, compensated friends-and-family test group. Capture vertical video clips during tests for creative learning.
Create marketing assets (days 7–10): Produce 3–4 short vertical assets per concept: teaser, behind-the-scenes, sensory close-up, and owner/chef story. Include QR code or unique promo code for tracking.
Soft-launch & seed (week 2): Post organically across platforms and push to local creators. Offer a limited “first 48 hours” discount to early redeeming customers tied to the unique code. Use micro-boosting to amplify best-performing asset.
Measure signals (weeks 2–4): Track video KPIs (watch time, completion, shares, UGC remixes), on-prem performance (redemptions, add-on rate), and local search spikes. Correlate the video-to-sales conversion with the unique codes/QR scans.
Iterate (weeks 3–4): Use Gemini to propose tweaks based on comment sentiment and sales patterns — smaller portion, spicier glaze, different plating — then release updated assets.
Decision gate (week 4): If conversion, margin, and UGC growth meet pre-set thresholds, scale the item for a longer run or permanent menu test. If not, archive learnings and pivot to the next concept.
Scale & institutionalize (weeks 5–8): Document SOPs, supplier changes, and predictive reorder quantities driven by anticipated demand uplift from video reach.

Practical templates: Gemini prompts you can use today

Use these prompts as starting points when working with Gemini-like models. Tweak for tone and local context.

Concept Synthesis Prompt: "You’re a menu strategist. Given these trends [list], produce five limited-time menu concepts that use existing line equipment, include one-sentence selling hooks, estimated plate cost, and a 30-second vertical-video creative brief."

Recipe Scaling Prompt: "Scale this single-plate recipe to a 50-plate batch for a busy lunch shift. Provide ingredient quantities, yield adjustments, three-line prep steps for cooks, allergen notes, and a rough food-cost percentage based on these local prices [list]."

Creative Brief Prompt: "Write a vertical-video script (0–15, 15–30, 30–60 sec versions) optimized for 2026 short-form platforms: include 3-second hook ideas, sensory shot list, recommended audio types, and 3 CTAs that drive immediate redemptions or UGC."

Video production: craft for watch time and conversion

Your LTO lives and dies by the first three seconds in 2026. Use these production rules:

Lead with sensory motion: steam rising, oil sizzling, a fork pulling cheese — not the chef talking.
Use a 3-shot cadence: hook (0–3s), reveal (4–15s), payoff or CTA (15–30s).
Native sound matters: platform-viral tracks or ASMR-level food sounds increase completion rate.
Localize quickly: swap one text overlay (neighborhood nickname, price) for geo-targeted promoting.
Make it shoppable: use QR codes and unique promo codes to link the video directly to redemption at the POS.

Seed your content with local creators — their followers fuel trust. Encourage UGC by offering a free side or drink for duets/stitches that tag your location and use a specific hashtag.

Measurement: what success looks like

Set both content KPIs and business KPIs before launch:

Content KPIs: completion rate > 50%, rewatch rate > 10%, remix rate > 1% of views, positive comment ratio > 70%.
Business KPIs: redemption-to-view conversion > 0.5% (local ads), add-on rate > 25%, break-even within test window (food cost + promotion vs revenue).

Decision rule example: If by Day 21 the item converts at 0.75% of localized viewers and yields a 20% contribution margin, scale. If below 0.25% conversion, sunset and harvest creative learnings.

Ops, supply chain, and compliance — don’t forget reality

Fast testing is valuable only if operations can absorb it. Before launch:

Confirm supplier lead times for specialty ingredients. If lead time > 10 days, rule out for a two-week LTO.
Run a cost sensitivity analysis: a 10% price shift in a top ingredient should not sink your margin.
Update allergen labels and staff training notes generated by Gemini.
Ensure data practices follow platform Terms of Service and privacy rules — use aggregated signals and platform APIs rather than scraping personal data.

Real-world example: Solstice Kitchen’s hot-honey calamari (hypothetical case study)

In late 2025, Solstice Kitchen — a 60-seat neighborhood bistro — monitored vertical-video data showing rising engagement for “crispy seafood + sweet heat” formats in their city. They used the 10-step playbook to execute:

Week 1: Trend scoring surfaced “hot-honey calamari” as a top cluster.
Week 2: Gemini generated 4 concept variants; they prototyped two and shot 5 short assets.
Week 3: Organic posts + two local creators resulted in 95k local impressions, 62% completion rate, and 420 unique QR redemptions in 10 days.
Outcome: The dish converted at ~0.7% of local viewers, delivered a 22% contribution margin, and became a weekend staple for 8 weeks — later joining the permanent small-plates menu after seasonal adjustment.

Lessons learned: rapid creative iteration and a low-friction QR redemption were crucial. The kitchen’s ability to scale a single prep step (single-batch batter mix) made the item sustainable.

Ethics, privacy, and platform considerations in 2026

Using platform signals responsibly is essential. Best practices:

Favor aggregate, anonymized signals over user-level scraping.
Respect platform rate limits and API terms of service.
Disclose when using influencers or paid promotions — transparency builds trust with diners.
Keep customer data (redemptions, emails) under strict consent frameworks; offer opt-outs for marketing pushes.

Future predictions: what comes next for AI menus and vertical-video insights

Looking forward from 2026, expect three converging trends:

Shoppable verticals become real-time signals: Platforms will continue to integrate commerce primitives that link views directly to inventory and POS in near real-time.
AI-driven supply chains: Predictive ordering informed by expected viral reach will allow restaurants to buy ahead without overstocking.
Platform-native dishes: Some LTOs will be designed for shareability first and palatability second; the winners will balance both.

Holywater’s move into AI-curated IP (Forbes, Jan 2026) and the continued rise of Gemini-style guided learning mean that the tools to do this affordably are now mainstream. Restaurants that adopt a data-first, rapid-experimentation culture will find LTOs to be a powerful growth lever.

Set a 4-week calendar with decision gates.
Pull 2 weeks of vertical trends for your metro area.
Use Gemini to create 5 concepts and 10 creative briefs.
Prototype 2 dishes, film 6 vertical assets, and seed local creators.
Use QR codes/unique codes to track redemptions.
Measure at Day 21 and decide: scale or shelve.

Final thoughts

The old model — long product cycles, big bets — is giving way to rapid experimentation fueled by vertical-video insights and AI. When restaurants treat short-form consumption data as a real-time market research panel and pair it with guided AI for rapid iteration, limited-time offers stop being random acts of culinary hope and become precision tools for growth.

Ready to prototype your first AI-generated menu item? Start by pulling a fortnight of local vertical-video trends, then run the Gemini concept prompt above. Need help operationalizing the playbook? Contact our team for a tailored workshop that turns your kitchen into a data-driven menu lab.

flavours

Contributor

Senior editor and content strategist. Writing about technology, design, and the future of digital media. Follow along for deep dives into the industry's moving parts.