A furnace tune-up ad turns up between a school photo and a weather post, and nobody in that feed went looking for it. That one fact explains most of what feels slippery about paid social, and it is why "how do Facebook ads work" is really two questions if you run a local business. The first is mechanical, about how Meta picks which ad appears in front of which person. The second is commercial, about whether that produces work you get to book and invoice.
Facebook ads buy the first of the six growth levers we work through with local owners: Attention, meaning whether the right people inside your service area know you exist and have a reason to raise a hand. Attention is the one lever a local business can purchase on demand, which is most of the appeal. It stays a cost until Conversion catches it and turns a raised hand into an appointment, so an impression is not a result and neither is a click. A form fill only starts counting once somebody answers it.
How do Facebook ads work, from audience to impression
Two separate decisions sit behind every ad you see. The first is eligibility: when you build a campaign you define who is allowed to receive it, including required controls like location, which produces a pool of people your ad can reach, though nobody has seen it yet.
The second decision happens in a fraction of a second. Somebody opens the app, a slot in their feed becomes available, and every ad eligible for that person and placement enters an auction for it. Meta scores each one, the ad with the highest total value takes the slot, and the auction runs again for the next slot, millions of times a day.
Response data from delivery feeds back into Meta's predictions, so the system's guess about who is likely to act keeps getting revised while your campaign runs. Your ad is not sitting in front of a list you handed over. It is competing, one impression at a time, for people the system believes are worth showing it to.
Why the highest bid does not always win
Meta's public explanation names three inputs into what it calls total value: your bid, its estimated action rate for the person in front of it, and ad quality. The closest compact version is that your bid gets multiplied by the estimated action rate to represent what the impression is worth to you, with an ad quality contribution added on top.
Estimated action rate is Meta's prediction that this particular person will take the action tied to your objective. Ad quality is a model-based assessment that folds in how people respond to the ad, including negative feedback and traits Meta treats as low quality, such as sensationalism or engagement bait.
Money alone therefore does not buy the slot, and an ad that is clear about what it offers, shown to people the system expects to respond, can beat a higher bid on total value. Meta names those three factors without publishing the production formula or the weights, so treat any precise-looking equation, including that one, as a conceptual model rather than a spec.
Predicted demand and expressed demand
Google Search starts with a sentence somebody typed. Google's own material treats a query like emergency locksmith near me as ready-to-book demand, matching a stated job to a provider who can do it now. Meta starts earlier, and its lead generation material describes an environment where the opportunity extends beyond people already searching, where a business can create demand rather than only capture it. There is no query, so eligibility and prediction stand in for one.
That gap decides what an ad is realistically able to do. A homeowner with no heat at 6 a.m. is typing into a search bar rather than scrolling, and a feed ad reaches them after the decision is already made. The same homeowner in September, seeing a clear ad for a tune-up with a photo of work they recognize, is a different opportunity entirely, because the demand did not exist yet.
Neither channel outranks the other in general, and the useful question is which of those two moments produces most of the work you sell.
Who Meta decides to show your ad to
Meta can draw on age, location, interests, activity across its own apps, interactions with pages and ads, and permitted signals you supply from your website or customer list. Custom Audiences represent people who already have some relationship with you, and Lookalike Audiences model similarity to that group. What you are building is a pool of eligible people rather than a hand-picked list of names.
Under Advantage+ audience, most of what you enter behaves as a suggestion. Meta prioritizes what you selected, then expands past it when it predicts better performance somewhere else. A short set of settings stays firm, and Meta names minimum age, location, language, and Custom Audience exclusions as controls that expansion should not cross.
For a business with trucks and a drive time, location carries the most weight of any setting, because it is one of the few boundaries the system will not reach past. Put your real coverage in there instead of a tidy radius.
What kind of job you are asking the ad to sell
Buyer occasion predicts fit better than industry does. Inside one HVAC business, an emergency no-heat call and a spring maintenance plan are two different products bought in two different states of mind, and only one of them can plausibly be created by an ad nobody went looking for.
Occasions that sit naturally with Meta tend to share two things: the demand can exist before a crisis, and the outcome is visible enough that a stranger will stop for it. A landscaping transformation photographs well and can be sold in September for April. Cosmetic dental work and new-patient education qualify for similar reasons, though the clinical side makes the path from interest to chair longer.
Public benchmark data hints at the same pattern without proving it. In WordStream and LocaliQ's 2025 U.S. sample, traffic campaigns averaged a 1.71% click-through rate across all industries, with shopping and gifts up at 4.13% and automotive repair down at 0.80%. Categories qualified for that study with as few as three active campaigns, so read the individual figures as directional. The spread still fits the idea that in-market volume varies by category.
If nearly every profitable job you sell starts with a burst pipe or an outage, Meta is more likely to build familiarity than to be your primary source of booked work, and that is a fit finding rather than a failure.
Four checks before you spend much
- Pull your last twenty won jobs and mark each one as urgent and query-led, planned or seasonal, referral or trust-led, or influenced by discovery. That is an afternoon of work with no money moving, and while twenty is a small internal sample rather than a statistical finding, the shape shows up fast.
- The Meta Ad Library will show you currently active ads from comparable local providers near you, which establishes that the channel is being used in your market but not that anybody is making money on it.
- Somebody outside the business should be able to glance at your proposed image and first line and then tell you what the service is and what to do next. Meta's creative guidance leans on those opening seconds, and the three-second limit we use for the glance is our own test rather than a published threshold.
- Your own form and your own phone number are the last check. Fill one in as a stranger would, then write down who received it, how long that took, and where a booked appointment and a sold job would eventually be recorded.
What early evidence can and cannot prove
Click-through rate is an Attention measurement. It says people found the ad interesting enough to tap, and says nothing about whether they were inside your service area or in any position to buy. A platform lead count is a Response measurement, which tells you a form was submitted rather than that a real customer is waiting on you.
Early numbers also move for reasons unrelated to fit. Meta says delivery is less stable while the system is still learning, and significant edits can send an ad set back into that state, which makes the opening days of a campaign the least reliable evidence it will ever give you.
Google published an analysis of its own July to December 2025 data using a thirty day click window, reporting that standard search campaigns captured about 70% of conversions inside that window while its demand generation campaigns captured about 40%. Those are Google's numbers about Google's campaign types rather than a Meta measurement, and the pattern still matters here: activity that creates demand tends to convert later, and a short measurement window will undercount it.
Where the money is actually decided
An ad can win its auction, reach exactly the right person, and still produce nothing. The break is almost never in the auction itself. It sits in the space between a stranger raising a hand and somebody in your shop calling them back while they still care.
That is the handoff from Attention to Conversion, running through Trust and Response along the way. Somebody who taps a feed ad has no relationship with you yet, so proof has to be visible fast, whether that is before-and-after work or a job they recognize from their own street. Then somebody has to answer, and we have written about what it takes to answer every call, text, and message without adding headcount.
Two other posts pick up where this one stops: how much Facebook ads cost for what the attention itself tends to run, and how to run Facebook ads without wasting money for protecting the budget once the fit looks plausible.
If you would rather not work it out alone, the Attention Generation Engine is how we build and run this side of the wheel, at $3,500 to build and $2,000 a month, with ad spend paid direct to the platforms. You can also start with a free consultation, where we will look at where your won work comes from and what a pilot would cover. Easy to start, easy to stop.
