Meta has published two engineering posts and briefed the press about changes to the system that decides which ads people see on Facebook and Instagram. If you advertise your business there, here is what changed in October 2026, in plain language, and what could be worth testing.

What changed

Three updates, with dates:

GenRecs, a new recommendation model (official engineering post, 19 October). Instead of tracking each ad by campaign IDs, language models describe every ad by concepts – the product, style, offer – using shared building blocks called semantic tokens. Because every ad is described in the same terms, the system can relate a brand-new ad to ads it already understands. Paired with Meta’s other ranking models, Meta reports this delivered a +0.35% lift in conversions in production (Meta internal data, measured February 2026).
An overhaul of Andromeda (official engineering post, 28 October). Andromeda is the retrieval system that shortlists ads for each person. Meta reworked it with Nvidia so more ad data sits on the GPUs themselves, and each ad is routed to specialised “expert” models.
A press briefing (28 October). AdExchanger, interviewing Meta’s VP of ads infrastructure Matt Steiner, reports that Meta’s ranking system now scores every eligible ad for every impression, at more than a billion ads per second. According to Meta, ranking improvements drove 8.3% more clicks and a 15.7% lift in conversions on Facebook in Q2. Those are Meta’s own platform-wide figures – useful as direction, not a promise of what any single account will see.

A hypothetical example

Imagine a new cafe in Boisar launches its first ads today, with zero clicks and no history. In principle, semantic tokens let the system relate those ads to similar cafes and restaurants people already engage with, so the ads may start with a more useful signal than pure trial and error. That is the benefit Meta is aiming for, not a guaranteed outcome. Steiner cautions that the learning phase shortens rather than disappears – new ads still have to earn real results.

What it means for advertisers

The direction is steady: Meta’s system does more of the matching work itself. Steiner’s summary: “The importance of targeting is diminishing over time as ranking models get better at selecting who an ad would be relevant for.” That puts more weight on the inputs advertisers control – the variety of creative you supply, and the quality of the conversion data you send back.

What could be worth testing

None of this requires rebuilding your account. A few measured tests:

Tighten conversion tracking. Running the Conversions API alongside the Meta pixel can recover signals the browser pixel misses – but only when events are deduplicated (matching event IDs, so the same lead is not counted twice) and set up within the consent rules that apply to your visitors. Installing both tags without deduplication can make reporting worse, not better.
Test creative variety. Try adding genuinely different concepts – different hooks, offers, formats – alongside your current ads, and judge on leads and cost per lead rather than clicks.
Test a simpler structure. On a slice of budget, run a broader or Advantage+ style campaign against your current setup, and keep whichever wins on business outcomes.

Our take

For small and mid-sized Indian businesses, the practical response is testing, not rebuilding. Ad matching is getting more automated; the levers that remain yours are clear offers, honest creative, and clean data.

Questions about how this affects your campaigns? Book a free strategy call and we will look at your account against these changes.

Sources

Want this working for your business?

Book a free 30-minute strategy call - we will audit your marketing and tell you what to fix first.

Book a Free Strategy Call

Meta Changed How Facebook and Instagram Ads Get Ranked – What Happened and What to Test