Mohammad Nasravi

Chat with Mohammad on WhatsAppگفت‌وگو با محمد در واتساپتواصل مع محمد عبر واتساب

AI Daily: Meta’s ad engine now reads your creative

Branded AI Daily card over a dark archive hall where a thin cyan beam sweeps across rows of glowing glass panels, with the headline in white and cyan.

Written by Mohammad Nasravi inنوشته شده توسط محمد نصراوی در

Meta’s ad system now uses language models to read the text and images in every ad and describe it by brand, colour, fit and style. Its own ads chief says targeting matters less each year. That makes the frame itself the targeting, and the brief has to be written for a machine that looks.

Meta’s ranking model now reads the ad itself

Meta announced three ranking upgrades on 28 October. The new one, GenRecs, uses LLMs to read an ad’s text and images and describe it in shared “semantic tokens” such as a product’s brand, colour, fit and style, so a new ad can borrow what the system knows about similar ones. Meta says ranking improvements drove 8.3% more clicks and 15.7% more conversions on Facebook in Q2. Matt Steiner, its VP of ads ranking, said “the importance of targeting is diminishing” and that the cold-start phase gets shorter, not solved: “You still have to learn” (AdExchanger, 28 Oct).

My read: If the model reads the frame, the frame is the brief. A pack half hidden behind a hand, a colour graded off-brand, a logo that only appears at the end: the machine files the ad under the wrong neighbours. On one launch I cut ten opening hooks from the same film and let the data pick; the winner beat my favourite. That test now starts earlier, inside the ranking. So put the product, its real colour and its name in the first seconds, clearly. Pretty but vague will be read as vague.

Instagram’s boss: AI content probably ranks worse

On the Colin and Samir podcast, Instagram head Adam Mosseri said anger about AI is “going to continue to get worse”, especially in creative industries. He said Instagram does not rank AI creators differently, then added: “Do we rank AI content as well as we rank non-AI content? Probably not.” He also said Instagram is starting to require AI accounts to label themselves (The Verge, 28 Oct).

My read: Two statements from the same company, the same week. The ad engine wants clear, readable creative; the feed is cool on content that looks generated. The answer is not hiding the tool. It is the cut. I throw away 60–70% of my takes, and most of what goes is the stuff that looks like every other AI clip: the drifting camera, the plastic skin, the glow. My Christmas greeting was made end to end with AI tools. It works because it was directed shot by shot, not because nobody can tell.

Runway App Studio turns a team’s prompts into apps

Runway launched App Studio on 26 October. Describe a tool in plain language and it builds an app with its own interface that generates images, video and audio from your Brand Kits and Workflows; an existing Workflow becomes an app “in one click”. Built-in apps include AutoComp, which Runway says swaps a product into existing footage “while keeping labels and packaging accurate”. It is included in every paid plan and runs on existing credits; Enterprise admins can switch it off or block public links (Runway, 26 Oct).

My read: Runway names the real problem: the knowledge sits in one person’s prompts. In-house, that person is usually me. An app can hand my settings to a brand manager. It cannot hand over the 60–70% I reject. On a full group range, the risk is an app producing forty SKU versions and nobody checking the pack in each one. If you build one, build the check into it: a named person approves every product frame before it leaves.

What I’d do today

Open your three live ads on Meta and pause on the first second. Can you read the product, its colour and its name without sound? If not, recut the opening before you change any audience setting. If you want a film built for that first second, I direct it shot by shot.

Sources

خبرهای هوش مصنوعی را در تلگرام دنبال کنخلاصهٔ خبرها و برداشتِ من از کارِ واقعی، در کانالِ Nasravi.com در تلگرام.
عضویت در کانال تلگرام

Written by Mohammad Nasravi — AI creative director, in-house at a major FMCG group. Four years putting AI-made commercials past real brand teams, including Shirin Asal campaigns for Gulfood 2026 and Canton Fair 2025. More about the work · Start a project نوشتهٔ محمد نصراوی — کارگردان خلاق هوش مصنوعی، درون یک گروه بزرگ FMCG. چهار سال است کارِ ساخته‌شده با هوش مصنوعی را از فیلتر تیم‌های برند رد می‌کند؛ از جمله کمپین‌های شیرین عسل برای گلفود ۲۰۲۶ و کانتون فر ۲۰۲۵. دربارهٔ کارها · شروع پروژه بقلم محمد نصراوي — مدير إبداعي للذكاء الاصطناعي من داخل مجموعة FMCG كبرى. أربع سنوات من تمرير إعلانات مصنوعة بالذكاء الاصطناعي عبر فرق علامات حقيقية، منها حملات Shirin Asal في جلفود 2026 ومعرض كانتون 2025. المزيد عن العمل · ابدأ مشروعك