[{"data":1,"prerenderedAt":458},["ShallowReactive",2],{"guide-facebook-ad-interest-research":3},{"id":4,"title":5,"answer":6,"authorId":7,"body":8,"category":349,"ctaVariant":350,"dataset":349,"description":351,"examples":352,"extension":353,"faqs":354,"heroImage":379,"intro":380,"meta":381,"navigation":382,"path":383,"publishedAt":384,"seo":385,"sources":386,"stats":418,"stem":456,"updatedAt":384,"__hash__":457},"blog\u002Fblog\u002Ffacebook-ad-interest-research.md","Facebook Ad Interest Research (2027 Method)","Research Facebook ad interests by treating them as hypotheses, not as filters. Brainstorm seed interests from your offer and your real customers, vet each one against the audience-size estimate and the Suggestions and Browse tools in Ads Manager, then validate the angle it represents against live ads in the free Meta Ad Library, where an ad running for months is probably converting. Since 23 June 2025 Meta has consolidated many granular interests and now treats detailed targeting as audience suggestions by default, with only location and minimum age as true hard constraints, so the deliverable is a small tested set of proven angles and a clean seed you hand to Advantage+, not a deep interest stack.","likit-sae-lee",{"type":9,"value":10,"toc":338},"minimark",[11,16,20,23,26,45,52,56,59,62,65,121,124,127,131,134,137,140,143,149,153,156,159,167,171,174,177,225,228,231,235,243,246,249,311,319,325,329,332,335],[12,13,15],"h2",{"id":14},"interests-are-seeds-now-not-fences","Interests are seeds now, not fences",[17,18,19],"p",{},"Old interest research had one move: open the detailed-targeting box, stack a dozen hobbies and brand pages you thought your buyer liked, and trust the fence to keep the wrong people out. That model is broken, and Meta broke it deliberately. The platform now treats what you type as audience suggestions by default, showing your ad to people who match and to others when it predicts a better result. On Meta's own Advantage+ Audience documentation, only location and minimum age remain true hard constraints (language and custom-audience exclusions are also respected, but the interest layer is not a wall). The picker still exists. It just no longer does the job people remember it doing.",[17,21,22],{},"The removals make the direction unmistakable. Meta began consolidating many granular interest categories in Ads Manager on 23 June 2025, merging niche segments tied to sports, food, music genres, car models and more into broader groups, and campaigns built on the retired options before that date were set to stop delivering after 15 January 2026 if left unchanged (Social Media Today, 2025). The long tail of surgical interests that power users leaned on is being pruned on purpose. Underneath it sits Andromeda, the ad-retrieval engine Meta detailed on its engineering blog on 2 December 2024, which reads signals in the creative itself (the visuals, the hook, the format) to match ads to people, and which Meta reported lifted retrieval recall by about 6% and ad quality on selected segments by about 8% (Engineering at Meta, 2024). When a machine that good is choosing the audience from creative signals, a hand-built interest wall mostly gets in its way.",[17,24,25],{},"So why research interests at all? Because a good seed interest still does two useful jobs. It teaches you which motivation to name in the creative, which is the thing the retrieval engine actually reads. And it gives a cold account, one with little conversion history, a sensible starting signal to hand the automation before it has learned. The work has not disappeared. It has moved from building a fence to generating angles and proving them.",[17,27,28,29,34,35,39,40,44],{},"Keep this page in its lane. Deciding who your buyer is in the first place is ",[30,31,33],"a",{"href":32},"\u002Fblog\u002Ffacebook-ad-audience-research","audience research","; deciding whether to run interests at all versus going broad is ",[30,36,38],{"href":37},"\u002Fblog\u002Fbroad-vs-detailed-targeting","broad versus detailed targeting","; how the targeting system uses whatever you feed it is ",[30,41,43],{"href":42},"\u002Fblog\u002Ffacebook-ad-targeting","how Facebook ad targeting works",". This guide is the narrow, practical middle: how to generate interest candidates, read the signals that survive, validate them, and test them so you keep the ones that earn their place.",[17,46,47],{},[48,49],"img",{"alt":50,"src":51},"A marketer at a laptop mapping clusters of related ideas on sticky notes, calm modern workspace, no text visible","\u002Fimages\u002Fblog\u002Ffacebook-ad-interest-research-mapping.webp",[12,53,55],{"id":54},"brainstorm-seed-interests-from-the-offer-and-the-customer","Brainstorm seed interests from the offer and the customer",[17,57,58],{},"Never start in the interest picker. Starting there means you brainstorm from Meta's vocabulary instead of your buyer's, and you end up with whatever autocomplete suggests. Start with two things you already own: the offer and the customer. A seed interest is not a guess about a hobby, it is a testable claim that a particular motivation predicts a buyer, and the good ones come from evidence, not from the dropdown.",[17,60,61],{},"Work the offer first. Ask plainly what problem it solves, what it costs, and who has already paid to solve that problem another way. A $19 accessory and a $400 machine do not share a buyer even inside the same category, and price sorts intent before any interest does. Then mine the customer you already have. Reviews, support tickets, refund notes and the messages people send you are full of the adjacent products they mention, the brands they compare you to, the publications they read, and the identity they are reaching for. Those specifics are your raw seed list, in the buyer's own words rather than Meta's taxonomy.",[17,63,64],{},"Sort the raw list into three buckets, because each bucket carries a different bet:",[66,67,68,84],"table",{},[69,70,71],"thead",{},[72,73,74,78,81],"tr",{},[75,76,77],"th",{},"Bucket",[75,79,80],{},"The bet it makes",[75,82,83],{},"Example for a home-espresso gear seller",[85,86,87,99,110],"tbody",{},[72,88,89,93,96],{},[90,91,92],"td",{},"Direct",[90,94,95],{},"People already in the category will buy",[90,97,98],{},"Espresso, coffee roasting, home barista culture",[72,100,101,104,107],{},[90,102,103],{},"Adjacent",[90,105,106],{},"People who buy the complements also buy this",[90,108,109],{},"Specialty tea, kitchen gadgets, cooking shows, cafe culture",[72,111,112,115,118],{},[90,113,114],{},"Aspirational",[90,116,117],{},"People buying an identity or upgrade will convert",[90,119,120],{},"Minimalist home design, \"third wave\" coffee brands, weekend-ritual content",[17,122,123],{},"Direct interests are the safe, obvious bet and usually the smallest room. Adjacent interests are where volume and surprise live: the person who follows premium cookware may be a better prospect than someone who already owns three grinders and needs nothing. Aspirational interests target the identity the purchase confirms, which is often what the winning creative angle turns out to be. Write each seed as a one-line hypothesis in the buyer's language (\"wants cafe-quality coffee at home without the barista skill\"), because that sentence is both the interest you will test and the hook you will write. Ten to twenty seeds across the three buckets is plenty. You are generating candidates, not committing budget, so favour breadth here and let the next three steps cut the list down.",[17,125,126],{},"One caution that saves money later: an interest and an angle are not the same thing. \"Coffee roasting\" is an interest label; \"tired of paying six dollars for coffee you could make better at home\" is an angle. The label points delivery at a rough pool; the angle is what actually self-selects the buyer once the ad is in front of them. Keep both columns as you brainstorm, because in 2026 the angle is the part that does the heavy lifting.",[12,128,130],{"id":129},"read-the-signals-meta-still-exposes","Read the signals Meta still exposes",[17,132,133],{},"Half the internet still links to Facebook Audience Insights as the place to research interests. Stop looking for it. Meta discontinued the standalone tool on 1 July 2021 and folded a much smaller version into Meta Business Suite Insights (Meta Business Help Center, 2021), so the deep affinity reports and \"people who like X also like Y\" charts people remember are gone. What remains is a scattered set of smaller signals, and used together they are enough to vet a seed list before you spend.",[17,135,136],{},"Start with the audience-size estimate in the ad set, the potential-reach figure that updates as you add or remove an interest. It is a rough range, not a promise, but it answers the one question that kills tests quietly: is this segment large enough to ever gather results. This matters because the audience underneath is enormous. Facebook ads could reach about 2.28 billion people in early 2025, roughly 39.4% of all adults aged 18 and over (DataReportal, 2025), so a seed interest that shrinks your estimate to a sliver is almost always a segment too thin to leave the learning phase. If adding one interest drops the estimate by an order of magnitude, that is a warning, not a sign of precision.",[17,138,139],{},"Then use the two features built into the detailed-targeting box itself. Suggestions surfaces related interests once you have added one, which is a fast way to widen a bucket with terms you would not have typed (add \"espresso\" and it may surface \"moka pot\" or \"coffee subscription\"). Browse lets you walk Meta's interest, behaviour and demographic tree to see how the platform has grouped things after the 2025 consolidation, which tells you whether the granular seed you wrote still exists as a targetable option or has been folded into a broader parent. If a seed has vanished from Browse, it was pruned, and a campaign relying on it would have stopped delivering anyway.",[17,141,142],{},"Finally, read your own audience data in Meta Business Suite Insights and your organic page analytics for the age, location and top-content signals of people who already engage. Treat that as directional rather than decisive, because engagers are not always buyers, but it is free and it corroborates or contradicts the seeds you wrote. The point of this whole step is triage: keep the seeds that are real and reasonably sized, drop the ones that are tiny or have been consolidated away, and carry a shorter, saner list into validation. You are not looking for a magic interest here. You are removing the ones that would waste a test slot.",[17,144,145],{},[48,146],{"alt":147,"src":148},"A close-up of a hand adjusting a slider control on a screen while a size estimate gauge shifts, warm light, no readable text","\u002Fimages\u002Fblog\u002Ffacebook-ad-interest-research-signals.webp",[12,150,152],{"id":151},"validate-the-angle-in-the-free-meta-ad-library","Validate the angle in the free Meta Ad Library",[17,154,155],{},"A shortlisted interest is still only a hypothesis, and the cheapest place to test it is not your ad account, it is the free Meta Ad Library. The Library is public and requires no login (Shopify, 2026), and for interest research it answers a sharper question than \"does this interest exist\": is the angle this interest represents one that advertisers are already paying to run. Remember, you are validating the motivation, not the checkbox, because the checkbox is a suggestion Meta will expand past anyway.",[17,157,158],{},"Run three passes. First, search by the problem or category the way a buyer would phrase it, not just a brand name, so you surface advertisers competing for the same motivation that you did not know existed (Shopify, 2026). If your aspirational seed was \"cafe-quality coffee at home\" and three brands you had never heard of are all running ads on exactly that promise, the angle just got stronger for free. Second, read a single competitor's whole set of live ads rather than one, because the spread reveals which motivations they address: a price angle for the sceptic, a ritual angle for the enthusiast, a gift angle for the holidays. You are reverse-engineering their interest map from their creative, which is far more honest than any audience they would describe.",[17,160,161,162,166],{},"Third, and highest leverage, use ad longevity as a free performance read. How long an ad has been running is a public proxy for whether it works: as Shopify puts it, a brand that has run a particular ad for a long time likely found it effective (Shopify, 2026). An ad live for three months or more is probably converting, because nobody pays to keep a loser alive, while an offer that appeared and vanished in a couple of days probably flopped, so you can cross that angle off before you ever spend. A searchable archive of ads, such as the one ",[30,163,165],{"href":164},"\u002Fblog\u002Fmeta-ad-library-guide","AdPlay.ai"," maintains, makes it faster to read a whole category at once rather than one advertiser at a time. What you are deciding at the end of this step is which two or three angles, and therefore which interest sets, are worth a paid test. If the market is not sustaining ads around a motivation, promote a different seed rather than betting budget on a hunch the Library already argued against.",[12,168,170],{"id":169},"stacking-versus-layering-two-operations-that-pull-opposite-ways","Stacking versus layering: two operations that pull opposite ways",[17,172,173],{},"Most interest mistakes come from confusing the two things the picker lets you do, because they look similar and behave oppositely. Get this straight and half your setup errors disappear.",[17,175,176],{},"Stacking is adding several interests in the same detailed-targeting box. It uses OR logic: a person qualifies if they match any one of them, so the audience gets bigger with every interest you add. Espresso OR coffee roasting OR moka pot reaches anyone in any of those groups. Layering is the \"Narrow audience\" (or \"Define further\") button, which adds a second condition someone must also match. It uses AND logic: the person now has to sit in the overlap of both groups, so the audience gets smaller. \"Interested in coffee\" AND \"also interested in kitchen gadgets\" reaches only the intersection. Stacking widens, layering narrows, and reaching for the wrong one is how an ad set ends up either impossibly broad or starved to nothing.",[66,178,179,195],{},[69,180,181],{},[72,182,183,186,189,192],{},[75,184,185],{},"Operation",[75,187,188],{},"Logic",[75,190,191],{},"Effect on audience size",[75,193,194],{},"2026 verdict",[85,196,197,211],{},[72,198,199,202,205,208],{},[90,200,201],{},"Stacking (interests in one box)",[90,203,204],{},"OR (union)",[90,206,207],{},"Grows the pool",[90,209,210],{},"Fine for grouping one theme; expect delivery to expand past it anyway",[72,212,213,216,219,222],{},[90,214,215],{},"Layering (Narrow audience \u002F AND)",[90,217,218],{},"AND (intersection)",[90,220,221],{},"Shrinks the pool fast",[90,223,224],{},"Rarely worth it; each layer starves delivery for precision the creative gives free",[17,226,227],{},"Here is the trap layering sets. Every AND condition multiplies constraints and can shrink the pool by an order of magnitude per layer, and in 2026 you are paying that cost for precision the system will not honour as a fence and the creative would have delivered anyway. Power users used to layer three or four conditions to feel surgical; today that mostly hands the optimiser a tiny room to search and pushes the ad set toward the learning-phase wall, where it cannot gather the roughly 50 optimisation events in 7 days it needs to stabilise (Meta Business Help Center, 2026). A thin audience plus a normal budget is structurally stuck.",[17,229,230],{},"The practical rule: use stacking, not layering, and keep each stack to one clean theme. Put the closely related interests that all express a single motivation in one box, so a win is readable, and if you want to compare two motivations, run them as two ad sets rather than merging them into one blob you cannot interpret. Reserve layering for the rare case where the overlap is the actual buyer and the market is large enough to survive the cut. And treat even a good stack as a suggestion you hand Advantage+, not a wall, because that is exactly how the platform will treat it.",[12,232,234],{"id":233},"test-interest-sets-and-keep-the-ones-that-win","Test interest sets and keep the ones that win",[17,236,237,238,242],{},"Everything so far produces hypotheses. Testing is what turns them into a shortlist you trust, and the discipline is the same as any clean experiment: change one variable, hold the rest constant, and wait for enough data. Build separate ad sets that differ only in the interest theme, give them the identical creative, equal budgets, and the same optimisation event. If the creative or budget varies between them, you have learned nothing about the interest, only about the confound you introduced. The deeper mechanics of running these cleanly live in the ",[30,239,241],{"href":240},"\u002Fblog\u002Ffacebook-ad-creative-testing","creative testing guide","; the point here is that the interest is the variable, so nothing else may move.",[17,244,245],{},"Then respect the learning phase. An ad set needs roughly 50 optimisation events within about 7 days to exit learning and settle into efficient delivery (Meta Business Help Center, 2026), and a result read before that is noise. This is where budget math decides the test before the creative does. Suppose a purchase costs you around $20 and you give each interest set $30 a day. That is about one and a half purchases a day, roughly 10 a week, well short of 50, so every set is structurally stuck in \"Learning Limited\" and none of them can prove anything. You have three honest fixes: raise the budget so the math can clear 50 (here, north of $140 a day per set), optimise for a more frequent event higher in the funnel like Add to Cart so volume accumulates, or run fewer sets so the spend is not scattered too thin. Decide the budget from the event math, not from what feels comfortable.",[17,247,248],{},"Judge on cost per result over at least two weeks, never on an early click-through spike or a single strong day. Neutral benchmarks give you a sanity check on the numbers: WordStream's 2025 data puts the all-industry Facebook cost per lead at $27.66, up about 21% year over year, with a Leads-objective click costing $1.92 on average, though both swing widely by vertical and offer (WordStream, 2025). Those are reference points, not targets, and your real ceiling is your own break-even. A worked read of a three-set test makes the decision concrete:",[66,250,251,267],{},[69,252,253],{},[72,254,255,258,261,264],{},[75,256,257],{},"Interest set (Leads objective)",[75,259,260],{},"Weekly events",[75,262,263],{},"Cost per result",[75,265,266],{},"Verdict",[85,268,269,283,297],{},[72,270,271,274,277,280],{},[90,272,273],{},"Direct (espresso, home barista)",[90,275,276],{},"61",[90,278,279],{},"$24",[90,281,282],{},"Keep: cleared learning, under target",[72,284,285,288,291,294],{},[90,286,287],{},"Adjacent (premium cookware, cafe culture)",[90,289,290],{},"58",[90,292,293],{},"$19",[90,295,296],{},"Keep and scale: best cost per result",[72,298,299,302,305,308],{},[90,300,301],{},"Aspirational (minimalist home, ritual content)",[90,303,304],{},"22",[90,306,307],{},"$41",[90,309,310],{},"Cut: never cleared learning, thin and pricey",[17,312,313,314,318],{},"Read that table the way you would read your own account. The adjacent set won, which is common, because the person buying the complement is often a fresher, hungrier prospect than the one already deep in the category. The aspirational set failed on volume before it failed on cost, so its real lesson is \"too thin to judge,\" not \"bad angle,\" and you might retest it inside a broader stack rather than discard the idea. Keep the sets that beat your target, retire the rest, and read the delivery report on the winners to see which age, location and gender actually converted, because that real breakdown is better research than any interest label. The winning audience then becomes a seed for a ",[30,315,317],{"href":316},"\u002Fblog\u002Ffacebook-lookalike-audiences","lookalike audience",", so Meta can find more people who resemble the buyers a proven interest already surfaced.",[17,320,321],{},[48,322],{"alt":323,"src":324},"A phone held in one hand showing a scrolling feed of short video ads, thumb paused mid-scroll, soft daylight","\u002Fimages\u002Fblog\u002Ffacebook-ad-interest-research-test.webp",[12,326,328],{"id":327},"build-a-living-interest-shortlist-not-a-one-time-list","Build a living interest shortlist, not a one-time list",[17,330,331],{},"The reason interest research feels like busywork is that most people treat it as a document they fill in once before launch and never reopen. That is exactly backwards for a system that reprices by the hour and prunes its own targeting options on a schedule. The advertisers who keep their costs down treat the shortlist as a living thing: a handful of proven interest themes, each tied to an angle that survived the Library and a live test, refreshed as the market moves.",[17,333,334],{},"Keep the loop small and repeatable. Re-mine fresh reviews and support messages each month for new language and new adjacent products, because the words your buyers use drift and so do the seeds worth testing. Re-scan the Meta Ad Library on a schedule to catch which angles competitors are newly sustaining and which they have quietly dropped, since a long-running ad that disappears is the market telling you an angle died. Retire interest sets whose cost per result has crept up as they saturated, and promote a fresh seed into the open test slot. And watch Meta's own changes: after the 2025 consolidation, a seed that worked last quarter may have been folded into a broader parent, so verify it still exists in Browse before you rebuild around it.",[17,336,337],{},"None of this fights the automation. It feeds it. A tested interest becomes a clean seed you hand Advantage+, and the angle behind it becomes the hook the retrieval engine reads to find the buyer, which is where the real leverage sits now that only location and minimum age are hard fences. The mindset that compounds is simple: interests are experiments, not settings. Generate them from real evidence, prove them cheaply in the Library, test the survivors against your cost per result, keep the winners, and let the rest go. Do that on a rhythm and you stop guessing at checkboxes and start running a short list of angles you have actually seen convert.",{"title":339,"searchDepth":340,"depth":340,"links":341},"",2,[342,343,344,345,346,347,348],{"id":14,"depth":340,"text":15},{"id":54,"depth":340,"text":55},{"id":129,"depth":340,"text":130},{"id":151,"depth":340,"text":152},{"id":169,"depth":340,"text":170},{"id":233,"depth":340,"text":234},{"id":327,"depth":340,"text":328},null,"neutral","A repeatable method to research Facebook ad interests that convert: brainstorm seeds, read the signals, validate the angle in the Ad Library, then test and keep winners.",[],"md",[355,358,361,364,367,370,373,376],{"question":356,"answer":357},"How do I do interest research for Facebook ads in 2026?","Work in four steps rather than guessing at checkboxes. First, brainstorm seed interests from your offer and your existing customers, in three buckets: direct (the category itself), adjacent (what the same person also buys or follows), and aspirational (the identity they are buying into). Second, vet each seed against the audience-size estimate and the Suggestions and Browse tools in Ads Manager, dropping the ones that are tiny or vague. Third, validate the angle each interest represents against live ads in the free Meta Ad Library. Fourth, test a few interest sets against each other with the same creative and equal budget, and keep the ones that hit your cost per result. The output is a short list of proven angles, not a deep stack.",{"question":359,"answer":360},"Do interests still matter now that Meta expands past them?","They matter as inputs, not as fences. Meta's Help Center now describes detailed-targeting inputs as audience suggestions by default, and only location and minimum age remain hard constraints, so the system will show your ad beyond any interest you pick when it predicts a better result. That does not make interest research pointless. A good seed interest still teaches you which motivation to write into the creative, and it still gives a cold account a useful starting signal to hand Advantage+. The work simply moved from building a wall to generating and testing angles.",{"question":362,"answer":363},"Is Facebook Audience Insights still available for interest research?","Not the standalone tool. Meta discontinued Facebook Audience Insights on 1 July 2021 and folded a much smaller version into Meta Business Suite Insights, so the deep interest-affinity reports people remember are gone. The signals you still get are the audience-size (potential reach) estimate in the ad set, the Suggestions and Browse features inside detailed targeting that surface related interests, the audience tabs in Meta Business Suite, and the delivery report from a small live test. Use those to vet and prioritise seed interests. Stop chasing the retired tool that half the internet still links to.",{"question":365,"answer":366},"What is the difference between stacking and layering interests?","They are two different operations with opposite effects. Stacking means adding several interests in the same detailed-targeting box, which uses OR logic: someone who matches any one of them qualifies, so the audience gets bigger. Layering (the Narrow audience or Define further button) adds a second condition someone must also match, which uses AND logic, so the audience gets smaller because a person now has to sit in the overlap of both groups. Stacking widens, layering narrows. In 2026 both are treated as suggestions rather than hard filters, so heavy layering usually just starves delivery without buying the precision it once did.",{"question":368,"answer":369},"How many interests should I put in a Facebook ad set?","Fewer than you think, and often none. Because Meta expands past your interests anyway, a huge stack does little except muddy what you are testing. A practical approach is one clean theme per ad set: a small OR-stack of closely related interests that all express the same motivation, so a win is readable. If you want to compare two motivations, run them as two ad sets, not one merged blob. On a cold account a single tight seed interest or a customer-list lookalike gives the system a starting signal; on an account with healthy conversion data, broad with strong creative frequently beats any interest set.",{"question":371,"answer":372},"How do I find interests that actually convert?","You cannot know from the picker alone, because the interest label does not tell you whether that person buys. You infer it, then prove it. Infer by deriving seeds from real customers and by checking the audience-size estimate so you are not testing a segment too thin to ever leave the learning phase. Prove it two ways: validate the angle in the Meta Ad Library (are advertisers sustaining ads aimed at that motivation for months), then run a live test and read cost per result. An interest converts when the ad set built around it clears roughly 50 optimisation events in 7 days at a cost your margin can carry, not when it feels relevant.",{"question":374,"answer":375},"How do I test which interest sets are working?","Treat each interest set as a hypothesis and change one thing at a time. Build separate ad sets that differ only in the interest theme, give them the identical creative, equal budgets, and the same optimisation event, then let each clear the learning phase (about 50 events within 7 days) before you read anything, because a mid-learning result is noise. Judge on cost per result over at least two weeks, not on a strong day or an early click-through spike. Keep the sets that beat your target, retire the rest, and fold the winning audience into a lookalike seed. Read the delivery report to see which real people converted.",{"question":377,"answer":378},"How do I use the Meta Ad Library to validate an interest?","The Meta Ad Library is free and needs no login, and it lets you check the angle behind an interest before you spend. Search the problem or category the way a buyer would phrase it, not just a brand name, so you surface advertisers competing for the same motivation. Read across a competitor's whole set of live ads to see which angles they run for which buyer. Then use ad longevity as a free performance read: an ad that has been live for months is probably converting, because no one pays to keep a losing ad alive. If the angle your interest represents is one advertisers keep running, your hypothesis is stronger before a cent goes out.","\u002Fimages\u002Fblog\u002Ffacebook-ad-interest-research-hero.webp","You type one interest into the detailed-targeting box, Meta suggests forty more, and none of them tell you which will actually find a buyer. Picking interests on a hunch is how a budget ends up spread across people who were never going to convert. This guide gives you a method instead: brainstorm seed interests from your offer and your customers, read the few signals Meta still exposes, validate the angle each interest stands for against real ads, then test sets against each other so you keep only the ones that win. In 2026 the output is not a stack of interests, it is a short list of proven angles and a clean seed for Meta's automation.",{},true,"\u002Fblog\u002Ffacebook-ad-interest-research","2027-05-05",{"title":5,"description":351},[387,391,394,397,401,405,409,412,415],{"label":388,"url":389,"year":390},"WordStream \u002F LocaliQ, Facebook Ads Benchmarks 2025","https:\u002F\u002Fwww.wordstream.com\u002Fblog\u002Ffacebook-ads-benchmarks-2025","2025",{"label":392,"url":393,"year":390},"Search Engine Land, Facebook ad costs jump 21% in 2025, but still beat Google","https:\u002F\u002Fsearchengineland.com\u002Ffacebook-ad-costs-jump-beat-google-461690",{"label":395,"url":396,"year":390},"Social Media Today, Meta Is Consolidating More of Its Detailed Ad Targeting Options","https:\u002F\u002Fwww.socialmediatoday.com\u002Fnews\u002Fmeta-removes-more-detailed-ad-targeting-options-facebook-instagram\u002F757856\u002F",{"label":398,"url":399,"year":400},"Meta for Business, Advantage+ Audience","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fads\u002Fmeta-advantage-plus\u002Faudience\u002F","2026",{"label":402,"url":403,"year":404},"Engineering at Meta, Andromeda Personalized Ads Retrieval Engine","https:\u002F\u002Fengineering.fb.com\u002F2024\u002F12\u002F02\u002Fproduction-engineering\u002Fmeta-andromeda-advantage-automation-next-gen-personalized-ads-retrieval-engine\u002F","2024",{"label":406,"url":407,"year":408},"Meta Business Help Center, Transitioning from Audience Insights to Meta Business Suite","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fhelp\u002F531965364451139","2021",{"label":410,"url":411,"year":390},"DataReportal, Essential Facebook Statistics and Trends","https:\u002F\u002Fdatareportal.com\u002Fessential-facebook-stats",{"label":413,"url":414,"year":400},"Shopify, How to Use the Meta Ad Library","https:\u002F\u002Fwww.shopify.com\u002Fblog\u002Fad-library-facebook",{"label":416,"url":417,"year":400},"Meta Business Help Center, About the Learning Phase","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fhelp\u002F112167992830700",[419,423,426,430,434,438,441,445,449,452],{"label":420,"value":421,"source":422},"People Facebook ads could reach worldwide in early 2025","2.28 billion","DataReportal, 2025",{"label":424,"value":425,"source":422},"Share of the global 18+ population Facebook ads can reach","39.4%",{"label":427,"value":428,"source":429},"Date Meta began consolidating detailed interest categories in Ads Manager","23 June 2025","Social Media Today, 2025",{"label":431,"value":432,"source":433},"How Meta now classifies detailed-targeting inputs","Audience suggestions by default","Meta for Business, 2026",{"label":435,"value":436,"source":437},"Recall improvement to Meta's ad-retrieval stage from Andromeda (Meta's own figure)","+6%","Engineering at Meta, 2024",{"label":439,"value":440,"source":437},"Ads-quality improvement on selected segments from Andromeda (Meta's own figure)","+8%",{"label":442,"value":443,"source":444},"Date Meta discontinued the standalone Audience Insights tool","1 July 2021","Meta Business Help Center, 2021",{"label":446,"value":447,"source":448},"All-industry Facebook CPC, Leads objective, 2025","$1.92","WordStream, 2025",{"label":450,"value":451,"source":448},"All-industry Facebook cost per lead, 2025 (up about 21%)","$27.66",{"label":453,"value":454,"source":455},"Optimization events an ad set needs to exit the learning phase","~50 within 7 days","Meta Business Help Center, 2026","blog\u002Ffacebook-ad-interest-research","9N0-UrIVw2ig1KgZ8IBtNyAhWM-gKcVOBf6RXozQ8MQ",1786093700194]