Ad personalization

Instagram’s Help Center lists possible inputs to ad personalization: accounts followed and posts liked, content posted or interacted with, Facebook activity when accounts are linked, Facebook information and interests, and activity on other websites and apps.1 Mosseri explained in October 2025 that an apparently coincidental product ad can follow an earlier website visit: an advertiser may share information about visitors and pay to reach them with ads.7 Meta’s Newsroom says it does not use sensitive topics discussed with Meta AI—religion, sexual orientation, political views, health, racial or ethnic origin, philosophical beliefs or trade union membership—for ad personalization.8 Mosseri denies that Instagram listens through the microphone to target ads, and Meta’s Newsroom says microphone use requires permission and active use of a feature that needs it.78

Instagram’s Help Center says indicating interest or disinterest in an ad further personalizes the ad experience, while Meta’s Newsroom describes Ad Preferences as controls over both the ads shown and the information used to inform them.14 The “Why am I seeing this ad?” tool explains an ad’s appearance.4 Meta’s Newsroom describes Accounts Center as its settings area for managing accounts and preferences across its apps; grouping multiple accounts there is optional, and people who keep accounts separate can still apply the same settings individually.9 In January 2023, Meta distinguished the existing option to see fewer ads about unwanted subjects from an option it was still exploring to request more ads about subjects of interest.9 For people choosing free access in the UK, Meta’s Newsroom says all advertising tools and settings remain available.4

In June 2021, Meta announced a test of shopping ads that would choose a destination based on shopping behavior, sending people to selected Shop products or a business’s website depending on where they were most likely to purchase.10 Meta distinguished that test from existing Shopping Custom Audiences, which help businesses find an audience, and product-tagged ads that send people directly to a Shop; special offers for selected shoppers were a future possibility, not an announced available feature.10 A Meta patent describes another proposed approach for social-feed images: varying purchase links for tagged products according to viewers’ demographic information, but the patent does not establish that Instagram offers this feature.11

Meta’s engineering team describes several approaches to personalized paid delivery, beginning with Andromeda’s selection of a smaller set of potentially relevant ads from a large pool before later systems decide what to show.12 To improve the models making ad recommendations, Meta’s engineering team describes GEM as a large model that passes what it has learned to other models, with reported improvements in advertising results on Instagram and Facebook.13 Meta’s engineering team describes GenRecs as considering ad content and people’s preferences together, using short, reusable meaning-based labels instead of a separate identifier for every item, and reports that it already improves ad predictions.2 Unlike those reported operational approaches, Meta’s engineering team calls Hierarchical Interest Representation an exploratory research area: it combines interactions with information about businesses and products to connect people’s inferred interests with offerings even when an item has received little or no interaction.14