Cheap leads that don't convert get blamed on the audience settings. Since signal loss, the settings matter far less than the sentence at the top of the ad. Ten years ago a media buyer's edge was the audience. You knew which interests, lookalikes and exclusions to stack, and the platform did roughly what you told it. That edge is mostly gone. The machine picks the audience now, and the main thing it uses to pick is the ad you give it.
Which means the most important targeting decision in your account is getting made by whoever writes the first line. In most companies at this spend, that person isn't in the room when targeting gets discussed.
Signal loss, by the numbers
One large survey of more than 500 US advertising and data decision-makers [1] found that about 95% expected signal loss and privacy legislation to keep coming. Roughly 87% of buyers said their ad costs were rising because of it, and 82% said it had already changed how their organizations were structured. Most were leaning harder on first-party data, with around 71% expecting to grow their own datasets, and about nine in ten had changed how they personalize ads.
Two years later, a recent industry outlook [2] added a new layer: about 73% of the buyers it surveyed said they're prioritizing content shaped for AI-generated answers. Part of your audience is now a model summarizing you for someone else.
Look at what those buyers did about it. They restructured teams. They built datasets. They changed tactics. Almost every response was about getting more data in. Very few were about getting better words out.
The pipe got narrower. The ad got louder.
The technician sees signal loss as a plumbing problem. The pipe that used to carry user data got narrower, so you build a new pipe: server-side events, clean rooms, first-party lists, contextual buys. All worth doing. I'd fund every one of them.
The strategist sees something the plumbing view misses. My own start was in copy, which is why the first line of the ad is where I look first. When the system can't see as much about the person, it watches what the person does with your ad, then goes looking for people who behave the same way. So the ad now carries the instruction along with the message. Whatever kind of person your words attract first, the system will chase at scale.
That changes who's doing the targeting. The line at the top of the ad now does more of it than any setting the media buyer touches. Write something vague, and the system learns from whoever responds to vague things, usually the cheapest clicks it can find. Write something that names exactly who it's for, and the system learns from people who saw themselves in it.
The pixel wentquiet. The wordsstarted talking.
The technician can rebuild every pipe in the building and still lose this, because the cleanest data in the world can't fix an ad that describes nobody. The strategist wins it with a sentence.
Compression Survival: is your buyer still there after two squeezes?
I've written before about compression: answer engines squeeze your positioning into a line before they repeat it, and only simple, specific positioning comes through intact. Ad delivery does the same thing to your creative.
Every ad goes through two compressions now:
- The platform compression. The system boils your ad down to a pattern of who responded. That pattern is your real targeting.
- The answer-engine compression. AI tools boil your brand down to a sentence when someone asks who solves their problem. That sentence is your real positioning.
Compression Survival is the test: after both squeezes, is your buyer still recognizable? Can the system tell who you're for from the people who responded? Can the answer engine say what you do in a sentence a buyer would act on? Both questions get answered by what you wrote, long before anyone opens a dashboard.
Your Signal Map, the first thing you fill in inside the iFix Lab, starts with that second squeeze: what AI says about you today.
The math: the worse-looking ad is 32% cheaper per customer (illustrative)
These are modeled numbers, built for illustration. Read them as a model, and expect your own account to differ. Same channel, same $20 CPM.
Ad 1, broad and clever. Wide appeal, strong thumb-stop. Click-through rate 1.8%, conversion rate 2.5%.
- Cost per click: $20 divided by (1,000 times 0.018), which is about $1.11.
- Cost per acquisition: $1.11 divided by 0.025, which is about $44.
Ad 2, specific. Names the buyer and the situation in the first line. Fewer people click: 1.1%. But the ones who click are the right ones, converting at 6%.
- Cost per click: $20 divided by 11, about $1.82.
- Cost per acquisition: $1.82 divided by 0.06, about $30.
Ad 2 looks worse on every early metric a dashboard shows. It's about 32% cheaper per customer. And there's a compounding effect the table doesn't show: every conversion from Ad 2 teaches the system to find more people like that converter. Ad 1 teaches it to find clickers. After a few weeks, the gap usually widens.
Scale it. At $5M a month, the difference between a $44 and a $30 acquisition cost is the difference between about 114,000 and 167,000 customers. Same budget. Same channel. Same settings. Different words.
That's the whole thesis in one illustration. A technician optimizing on click-through kills Ad 2 in week one. A strategist reading the conversion line keeps it and writes three more like it.
Bars are scaled within each pair. The ad that loses on click-through wins on cost per customer.
What a targeting line looks like
Take a business software company selling to finance teams. A typical first line reads something like: Close your books faster with smarter automation. Everyone in finance agrees with it. Nobody in particular is described by it. The system will find whoever clicks on agreeable things.
Now rewrite it around a buying moment: For the controller who's rebuilding the same reconciliation spreadsheet on day three of every close. A CEO scrolls past. A junior analyst scrolls past. The controller who lived that exact Tuesday stops, and when that person converts, the system has a far sharper picture of who to find next.
That second line is doing the job an audience setting used to do. It names a role, a moment and a pain, in words the buyer would use about their own week. It also gives an answer engine something concrete to repeat.
Neither line isclever. One of themtargets.
The same move works in any category. Write the first line for the one person in the one moment, and let the portfolio of concepts supply the breadth.
Four ripples from writing for the machine
- Click-through rate becomes a misleading creative metric. Specific ads filter people out, so they often post lower click-through and higher conversion. Teams that judge creative on click-through kill the ads that target best.
- Concept diversity becomes audience diversity. Thirty variations of one idea teach the system one pattern. Ten genuinely different buyer situations teach it ten. At scale, the number of distinct messages sets the number of distinct audiences you can reach.
- First-party data and creative feed each other. Your customer lists tell you which buyer situations convert. Your copy then describes those situations. The system finds more of them. The loop only works if someone connects the data team to the writers.
- Rising costs hit vague accounts hardest. When buyers say costs are rising, they're rising unevenly. Accounts whose copy gives the system clear instructions waste less of every price increase.
Why going broad breaks between $1M and $10M a month
You'll hear “go broad and let the algorithm work” from almost everyone. At $1M a month with a few strong ads, that's often right.
Between $3M and $10M it breaks, because broad targeting with one dominant message teaches the system one narrow audience, then exhausts it. Teams see costs climb, nobody can pin down the cause, and everyone concludes the market's saturated and pulls back. Often they've only saturated the one buyer their copy described.
Same with “hook first, qualify later.” A scroll-stopping hook that attracts the wrong person is a targeting error that costs you twice: once for the click, and again for every week the system spends learning from the wrong people. Call it what it is: a sentence that said the wrong thing to the wrong person, at scale.
Past $10M a month, the concept roadmap is your targeting plan
- Creative strategy is audience strategy. The concept roadmap is your targeting plan. Budget it like one. Each new buyer situation you put into words is a new audience the system can find.
- The creative brief needs a who-it's-for line the system could act on. Skip the persona deck and write one sentence describing a real buying moment, specific enough that the wrong person scrolls past.
- Answer-engine presence becomes a writing deliverable. If AI tools describe you wrong, your paid creative is working uphill against a bad summary. Somebody needs to own how the brand compresses, across ads, site and public answers, and that somebody should be able to write.
- Data, creative and media teams need one meeting. At this spend, the handoff between the people who know which customers are worth most and the people writing ads is where most of the targeting value leaks. Whether you're the CMO, the in-house growth lead or the agency owner running the buy, put them in the same room on a schedule.
- Clean rooms and partner data should feed the brief as well as the buy. Whatever you learn about high-value segments should land in the next round of copy within weeks, where the writers can use it.
Four questions before any concept gets more budget
Score every live concept on four questions before it gets more budget:
- Who does this ad tell the system to find? Write the answer in one sentence. If two people on the team write different sentences, the system's confused too.
- What does it filter out? A strong ad turns the wrong people away on purpose. If nobody would scroll past it, it isn't targeting.
- Which buyer situation does it own? Map concepts to situations. Gaps on the map are audiences you aren't reaching. Duplicates are money spent teaching the same lesson twice.
- Does it survive the sentence test? Read the ad and ask what an answer engine would say about you if this were all it knew. If that sentence wouldn't make a buyer act, the ad's working against your positioning.
Questions senior buyers ask
Since signal loss, is interest and audience targeting dead?
It's much weaker than it was. Exclusions and first-party seed lists still help. The ad does most of the work, and it does it whether you plan for that or not.
How many distinct concepts does a $10M-a-month account need?
More than most run. The right number depends on how many real buyer situations exist in your market. Map those first, then count the concepts that speak to each.
Won't specific ads cap our reach?
Each one reaches fewer people. A set of specific ads, each owning a different situation, usually reaches more of the right people than one broad ad does. Breadth comes from the portfolio.
How would I know the ads are even attracting the right client?
Watch new-customer quality by ad: return rates, repeat purchase, sales-call fit. When an ad's buyers look different from your best customers, it taught the system the wrong lesson.
What does content shaped for AI answers mean for paid teams?
It means your ads, your site and your public answers should describe the same buyer in the same plain words. Mixed descriptions compress badly.
Who should own the first line of every ad?
Someone who understands the buyer well enough to describe their Tuesday, and who sees the conversion data weekly. If those are two different people, introduce them.
Ten sentences
Take your top ten ads by spend and write, for each one, the sentence describing who it tells the system to find. Put the ten sentences side by side.
If they mostly say the same thing, you've found why your costs keep climbing, and no new pipe will fix it. Send the page to your creative lead before the next sprint. If you want to read your account this way every week, that's what the iFix Lab is for. On day one, Your Signal Map puts on one page where you get found today, what AI says about you and what attention costs you, so you can see whether your words survive both compressions. The weekly Signal Briefing keeps it current. Signal loss won't wait for your next planning cycle. Join the iFix Lab for $97 a month. Cancel any time, and month one carries a 14-day refund. Next note: the math behind creative fatigue at real volume, and why cheap AI variants don't fix it.