Social advertising, how the delivery system and the creative decide who sees your ad, and the two readings that say scale or stop

Social advertising is buying a placement inside somebody else private feed, and three things decide who actually sees it: the objective you choose, the creative you upload, and whether the campaign has enough conversion data to leave its learning stage. The audience boxes matter least of the three. On Meta, TikTok and LinkedIn the delivery system now finds the buyer from the creative and the conversion signal, so the practical work is to send a clean signal, give the system several genuinely different creatives, and then read two numbers before changing anything.

Harvested against the India geo target on its own on 22 October 2026, social advertising draws 1,600 searches a month in India, and against the United Kingdom target on its own on the same day it draws 210 a month in the United Kingdom. Those are two separate readings of two separate countries, never added together.

What a feed auction is actually choosing

A search auction has a query. A feed auction has none. Nobody asked for your ad, so the platform is choosing between your ad and several hundred others for one slot in front of one person who came to look at something else. It picks on a blend of your bid and its own estimate of how likely that person is to do the thing your objective names.

That second half is why two advertisers with the same bid pay very different prices. If the system believes your ad will earn the action, it needs less of your bid to win the slot. The lever you hold over that estimate is not the bid. It is whether the action is being reported back to the platform accurately, and whether the creative looks like something that person stops for.

Why the creative carries the targeting now

Interest targeting was once the sharpest tool on the panel. It is now the blunt one, because the audiences are enormous and the system is better at finding buyers inside a broad pool than a planner is at describing them in advance. Broad targeting plus five distinct creatives beats six narrow audiences with one creative between them, and it does so because each creative is itself a filter: the person who stops for a price comparison is a different buyer from the person who stops for a before and after.

So treat creative as the audience plan. Write down the five buyers you want and make one asset per buyer, with the first three seconds doing the sorting. For which platform suits which kind of targeting in the first place, our comparison of what each advertising platform lets you target is the place to start, because the answer differs sharply between a professional network and a short video feed.

The learning stage, and why a small budget never leaves it

Every delivery system needs a run of recent conversions inside one ad set before its estimates settle. Until then it is exploring, and the cost per result swings widely from day to day. Two habits keep campaigns stuck there permanently.

• Splitting a small budget across many ad sets. Four ad sets each getting a quarter of the money each collect a quarter of the data, so none of them settles. One ad set with the whole budget settles in a week.
• Editing during the exploration. A change to the budget, the audience or the creative restarts it. Decide what you are testing, then leave it alone for the agreed window.

If the conversion you optimise toward happens only a handful of times a month, optimise toward something earlier and more frequent, such as a qualified form start, and keep the sale as the number you report.

Three audiences, and nothing more

Most accounts carry too many campaigns for the money in them. Three stages cover almost every business.

• People who have never heard of you. Broad targeting, the widest creative variety, judged on cost per new lead or new customer rather than on clicks.
• People who visited and left. Built from site visitors and video viewers. Different message: the objection, the proof, the price.
• People who bought. The cheapest revenue in the account, and the stage most often missing. One campaign for the next purchase and one for the referral.

This is the push half of the plan, where you choose the moment rather than waiting for it. The split between push and pull advertising decides your campaign type and your creative before you open the ads manager, and getting it the wrong way round is the most common reason a feed campaign reads as an interruption.

The two readings that say scale or stop

Daily cost per result is noise. Two readings carry real information, and both need a window rather than a day.

1. Cost per result over a trailing seven days, against your allowable figure. Your allowable figure comes from your own margin, not from a published benchmark. Work out what a customer is worth to you over the first year, decide what share of that you will pay to acquire one, and that is the line. Under the line for seven days running is the signal to add budget, in steps of about twenty per cent so the campaign does not re-enter exploration.
2. Frequency inside the audience you are paying for. When the same people are seeing the same ad many times a week, the cost per result rises for a reason no bid change fixes. That is a creative supply problem, and the answer is new assets rather than a new audience.

Read them weekly, write the decision down, and keep the record. Three months of weekly readings tells you more about your own account than any industry average does.

When paid social is the wrong purchase

Two situations come up often enough to name. If your product is bought at a moment of need that the buyer initiates, such as an emergency repair, search advertising reaches that moment and a feed does not. And if your site cannot report a conversion back to the platform, buy the tracking work first, because every number above depends on it.

Questions about buying social advertising

How much budget does a feed campaign need to produce a usable reading

Enough to collect a meaningful run of conversions in one ad set inside a week. Work backwards from your own cost per result: if a lead costs you about what a day of your current budget buys, you are testing one lead at a time and the readings will swing. Consolidate into one ad set until the weekly figure steadies.

Should I run ads from the brand account or build a separate one

Run them from the brand account. The follower count and the post history are part of what makes the ad credible, and a fresh account with no activity behind it converts worse for the same spend.

How many creatives does one campaign need

Five that differ in their first three seconds, not five crops of one image. Replace the weakest one each fortnight and keep the winner running until frequency inside its audience starts climbing.

Is a boosted post the same as a campaign

No. Boosting takes an existing post and buys reach for it against a simple objective. A campaign lets you choose the conversion event, the audience stages and the placements. Boosting is useful for testing which organic post people stop for, and the winner then becomes a creative inside a real campaign.

What does the platform report that I should not put in a board pack

Reach, impressions and engagements. They move with budget and tell a finance reader nothing about whether the money came back. Report cost per new customer and the revenue behind it, and keep the platform metrics for diagnosing a campaign rather than for proving one.

The feed you cannot buy a placement in

Everything above assumes the buyer scrolls, sees something, and clicks. A growing share of the same buyers now opens an assistant instead and types the question they would have scrolled past an answer to. There is no auction in that answer, no bid, no placement and no creative. The assistant names two or three companies, cites a handful of pages it read, and the people it names did not pay for the slot. At AI Knows Us we take that reading across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews and date it, so a brand that has bought every feed placement in its category can also find out whether it is in the answer where nothing is for sale. Our own measurement of what the assistants actually cited across 597 readings found that 85.6 per cent of citations went to a vendor or agency own site, which is a different distribution of attention from any feed.