The practical response is not to stuff more keywords into captions. It is to make each brand video a content unit that a viewer can describe, a production team can differentiate, and a marketing team can test.

When people can tell a recommendation system—in their own words—what they want to see more or less of, a broad lifestyle film has a more visible weakness: it may look polished without making clear whose specific interest or task it serves.

This does not mean platforms have disclosed their ranking formulas. It does not mean a topic label guarantees organic reach. It changes the minimum question in a video brief. Beyond “does this look good?”, a brand should ask: “Could the right viewer accurately say why they want the next episode?”

What changed—and what has not been proved

On October 13, 2026, Meta announced Facebook’s “Your algorithm” tools for Feed and Reels. People can add and manage topics they want to see, choose “Interested” or “Not interested,” and provide more detail in natural language. Meta says recommendations can adapt in real time during the same session. The announcement applies to Facebook’s own rollout; it is not evidence that every platform uses the same inputs or weights.

The direction is not isolated. Instagram previously introduced recommendation resets and topic controls. TikTok launched Manage Topics in 2025, allowing people to dial the frequency of more than ten topic categories up or down in the For You feed. Together, these products show that explicit audience preference is becoming a more visible input. They do not prove that explicit text replaces watch time, shares, creator affinity, safety rules, novelty, or the many other signals used in recommendation.

Meta’s engineering work offers a more useful clue. In January 2026, its Facebook Reels team described a User True Interest Survey model trained with direct answers to the question, “How well does this video match your interests?” Meta reported an A/B test with more than 10 million users and improvements in survey ratings, engagement, and retention-related measures. Those are Meta’s own results, not an independent benchmark, but they confirm that the company is using direct perception feedback alongside behavioral data. The same paper also says interest matching includes audio, production style, mood, and motivation—not topic alone. That distinction matters for creative production.

So a video labeled “camping” is not automatically relevant to someone interested in lightweight weekend camping. The subject, demonstration, pace, creator voice, and reason to keep watching still have to agree.

Buy interest units, not a stack of generic lifestyle cuts

Consider a portable power brand commissioning ten “outdoor lifestyle” videos. Every cut contains a tent, a sunset, a product close-up, and a slightly different hook. The brand receives ten files, but may have bought only one vague content hypothesis.

Now reorganize the same production around ten distinct interest-task combinations: allocating power during a weekend car-camping trip; managing batteries for location photography; reducing pack weight for a short hike; or prioritizing essential devices during an outage. Each piece still belongs to the same brand world, but it answers a different question that a real person could articulate.

The second approach is not about gaming an algorithm. It creates differences that can be observed and discussed. A commissioning document should therefore specify more than “ten 15-second vertical videos.” For each video, it should identify an interest hypothesis, the viewer’s immediate task, the product proof, and a review ID. File count measures output. Interest hypotheses measure learning.

The ONCE SIGNAL test for interest-readable video

SIGNAL is a production and commissioning framework, not a platform score. It keeps brand, production, creator, and social teams working from the same content promise.

S — Subject: can the topic be said in one natural sentence?

Use language a viewer might actually choose. “Better living” and “freedom to explore” are too broad to distinguish one unit from another. “Storage for a small rental kitchen” or “a first long road trip with a dog” is specific enough to guide a script without forcing the same words into every caption.

I — Intent: what is the viewer trying to do now?

One interest can contain several motivations: learn, compare, be entertained, verify a claim, or decide what to buy. “Understanding coffee grind size” and “comparing how two grinders are cleaned” both sit inside coffee content, but need different shots and different evidence.

G — Grounding: why does the brand deserve to enter the topic?

The piece needs a visible action, design detail, process, constraint, or informed point of view—not a product cameo wrapped around a trending topic. Facebook’s March 2026 originality guidance says minor edits, simple compilations, and content without meaningful new value may be deprioritized, while genuinely new information, analysis, or substantial narrative improvement can qualify as original transformation. That guidance is another reason to commission evidence, not cosmetic variation.

For a brand, the most defensible original value is often something its product, people, or actual use conditions can uniquely show.

N — Narrative: what job does the episode do in the series?

Mark each unit as an entry point, explanation, proof, objection, comparison, or next step. A series is not one script with ten cosmetic edits. It is an ordered set of questions around a durable interest. If one episode underperforms, the team can then examine the subject, intent, evidence, or expression instead of declaring the whole content pillar a failure.

A — Alignment: do picture, sound, caption, creator, and destination agree?

If the opening promises “storage for a small home,” the middle becomes a generic beauty montage, and the landing page talks only about premium materials, the viewer receives three different promises. Review the first frames, spoken or written language, post description, creator delivery, and destination for semantic alignment. Alignment does not mean verbatim repetition; it means every surface supports the same reason to care.

L — Learning: what reusable record survives publication?

Give each interest unit a unique ID and record its subject, intent, proof, version differences, account, rights, and evaluation window. Watch time, shares, comments, clicks, and conversions answer different questions. One high metric should not be promoted into a causal conclusion about audience interest. A future brief should begin with patterns across multiple units, not the visible style of one breakout post.

Change the cost denominator

“Cost per video” is total spend divided by exported files. It is useful for capacity planning, but it cannot tell a marketing lead how much valid learning was purchased.

A better denominator for an always-on program is the number of completed interest units. A unit counts only when it has a distinct hypothesis, a recognizable proof difference, an identifiable version record, and a completed publication and review window.

Cost per completed interest unit = strategy, production, versioning, publishing coordination, and review cost ÷ interest units that meet the agreed completion rule

If a shoot exports 20 files that repeat two interest hypotheses, the learning denominator remains closer to two than 20. Conversely, a production that shares talent, lighting, and locations can expand the number of genuine questions and demonstrations without increasing shoot days at the same rate.

There is no universal price benchmark here. Product readiness, talent, locations, language, usage rights, market count, and version scope all change the budget. The purpose of the denominator is to stop cheap duplicates from looking like broad learning.

Divide responsibilities before the camera rolls

The brand team approves claim boundaries, target markets, interest territories, and business destinations. The social or media team supplies account context, publishing cadence, metadata requirements, and evaluation windows. The production team turns each hypothesis into observable action, performance, sound, pace, and versions. It should not discover after the shoot that ten planned videos are one idea wearing ten hooks.

A useful brief can begin with six columns: interest subject, viewer task, brand proof, series role, version difference, and review ID. Those columns should then map into the shot list and delivery schedule. Brands building an ongoing program can connect this approach to a short-form production plan rather than treating each post as an unrelated mini-project.

When to adopt this model—and when not to

Interest-unit commissioning is useful when a brand publishes every month, one product has several use conditions, creator content and paid media need to share source assets, or the team already has comments, search terms, customer-service questions, and community discussions from which to form hypotheses.

It is unnecessary to rebuild a production system because of one platform feature if the brand makes one flagship film a year, serves an extremely narrow known audience, or has not yet settled its product facts and primary reason to buy. In those cases, clarifying the core claim and filming persuasive proof matters more than elaborate tagging. Regulated categories should also complete claim and compliance review before optimizing topic coverage.

Meta’s update matters not because brands can finally “read the algorithm,” but because viewers are gaining more direct ways to state what they want. Before the next video order, try a plain test: remove the brand name and logo. Can the team still describe, in one specific sentence, the interest and task the video serves? If not, more versions will only reproduce ambiguity faster.

Sources and scope

SIGNAL, the cost denominator, and the responsibility map are ONCE analytical tools, not Meta or TikTok operating instructions. Platform announcements support the stated product and research scope; they do not prove an industry-wide transition or guarantee reach, ranking, or commercial results.

← How Product Ads Make Humor Work Before You Blame the Video: Google’s Data Strength Changes Creative Evaluation →