Bring back the visitors worth bringing back,
and stop chasing the rest.
Someone read three pages, checked your service area and left without calling. That person is worth reaching again. The visitor whose browser is actively telling you to stop is not, and in California that request is a legal instrument. We build audiences from behavior you can account for, and we honor the signal your ad stack has probably never been wired to read.
Built for home services contractors, franchise systems and private equity portfolios — one location or two hundred. Run by an AI-first agency: AI-assisted audience modelling, and remarketing that reinforces the entity AI platforms and search engines draw on when a homeowner asks who to call.
Remarketing runs inside an AI-first program here. AI-assisted audience modelling decides who is genuinely worth reaching again, machine-verified conversion signals keep automated delivery honest, and the repeat exposure reinforces the brand entity AI platforms and search engines weigh when a homeowner asks an assistant which company to call.
The browser is telling you to stop. Your tag has never been asked to listen.
A visitor turns on a privacy setting. Their browser sends that preference with every request, including the one to your site. It arrives. And then the remarketing tag fires anyway, because nobody ever connected the thing that receives the signal to the thing that acts on it.
The mechanic is worth understanding because it is simpler than it sounds. Global Privacy Control is a setting a person turns on in their browser or extension. Once enabled, the browser sends the signal to the websites they visit, and participating sites respect it. There is nothing to install on the visitor's side beyond the setting itself, and nothing exotic on yours to receive it.
In California it carries legal weight. Civil Code section 1798.135 requires a business that sells or shares personal information to provide a clear and conspicuous link titled "Do Not Sell or Share My Personal Information" — and then provides that a business need not do so if it instead allows consumers to opt out through an opt-out preference signal sent with the consumer's consent by a platform, technology or mechanism, based on technical specifications set out in regulation.
Read that structure carefully, because it is the whole argument. The statute treats the signal as a real exercise of the right, equivalent to clicking the link. So a business that neither carries the link nor honors the signal has not chosen a lighter option. It has chosen neither.
The failure is almost never deliberate. A consent banner gets installed and receives the signal correctly. The remarketing tag gets installed separately, often by a different person in a different year, and fires on page load regardless of what the banner decided. Both components work. Nothing errors. The visitor's request arrives and is quietly discarded at the join between two systems nobody owns.
This is the same shape of failure we describe on website maintenance — everything returns a healthy result and the outcome is still wrong. The difference here is that the thing being discarded is a person's stated preference, which makes it a different class of problem from a broken form.
What we do about it is unglamorous. We check what the site receives, we check what the tag does with it, and where the two are not connected we connect them before touching anything else in the account. That is a morning of work, once, and it is the difference between a remarketing program you can defend and one you have simply never examined.
One list you can account for. One you inherited.
Audience size is the number every platform shows you and the number that matters least. What matters is whether you can say, for any record in the list, how it got there and whether it should still be there.
Nobody can say how it was built
- One list called "all visitors". Everyone who loaded any page, including people who bounced in four seconds.
- Membership duration left on the default. People who solved their problem months ago are still being paid for.
- A customer list uploaded at some point, by someone, under terms nobody recorded.
- No suppression. People who already bought keep seeing ads asking them to buy.
- Preference signals not connected, so people who asked to be left alone are in the list too.
Every record has a reason
- Lists built from meaningful behavior — a service page read, a form started, a call that did not book.
- Duration set to the decision window for that service, rather than to whatever the platform suggested.
- Customer lists used only where the terms allow it, with the basis recorded before upload.
- Buyers suppressed, because the ad budget belongs on people who have not decided yet.
- The preference signal wired through, so a request to stop actually stops something.
All four are published. All four are usually left on default.
Two come from law, two from the platform's own rules. None of them is obscure, and none of them is decided for you by the account setup wizard.
The statute names an opt-out preference signal, sent with the consumer's consent by a platform, technology or mechanism based on specifications set in regulation, as a way for a business to allow consumers to opt out of the sale or sharing of their personal information.
Receiving it and honoring it are different achievements. A consent banner that reads the signal correctly does nothing at all if the advertising tag next to it was never told to check what the banner concluded.
- Trace the signal from arrival through to the tag and connect the gap where one exists.
- Test it as a visitor would, rather than assuming the plugin description is accurate.
Google's restricted targeting policy sets out categories where personalized advertising is limited or prohibited, on the basis that building audiences around them is intrusive regardless of intent.
This bites in medical and financial verticals more often than operators expect, and it is worth establishing before an audience strategy is designed rather than after a disapproval.
- Check the vertical against the restricted list before building any audience around a condition or circumstance.
- Design around the restriction rather than testing where the edge is.
Uploading a customer list into an ad platform is a data transfer with conditions attached. Google's Customer Match policy governs what may be uploaded and on what basis.
The practical question is one nobody asks before the upload: what did these people agree to when you collected their details, and does that cover this? The answer lives in your own records rather than in the platform.
- Establish the basis before the file moves, and record it alongside the audience.
- Where the basis cannot be established for part of a list, that part does not go.
Google documents how data segments are built and how long a person remains in one. That duration is a setting, and the default is rarely the right answer for a service with a short decision window.
Somebody who needed a furnace repair in March has either solved it or moved on. Continuing to pay to reach them in June is not persistence, it is a setting nobody revisited.
- Set duration to the real decision window for that service, service by service.
- Suppress people who converted, so the budget stays on people still deciding.
Five links between a browser setting and a tag that fires.
The signal has a path, and every link in it is somebody's responsibility. On most sites four of the five work correctly, which is exactly why the failure survives — nothing is broken enough to notice.
What has to connect
The visitor enables the setting in their browser or extension. The browser transmits it with the request. Your site receives it, usually through a consent management layer. That layer decides a consent state. And finally the advertising tag either checks that state before firing, or does not.
The break is almost always at the fourth link. Consent layers are installed to handle banners and cookie categories, and they usually do that well. Advertising tags are installed by whoever set up the ad account, frequently before the consent layer existed. Neither party did anything wrong; nobody joined them up, and no error is produced when they stay separate.
Testing it is straightforward and almost never done. Enable the signal as a visitor would, load the site, and observe whether the tag fires. That takes minutes and it is the only evidence that matters — a plugin's feature list is a description of intent, not a measurement of behavior.
Beyond California, the picture varies by state and continues to move. We check the states you actually operate in rather than assuming one rule covers the country, and we treat the strictest applicable position as the working default because maintaining several behaviors is how inconsistency gets introduced.
- Browser setting — the visitor enables it once
- Request header — sent automatically with every visit
- The site — receives it, usually via a consent layer
- Consent state — the layer reaches a decision
- The advertising tag — checks that decision, or ignores it
- Your test — the only proof the chain is joined
Fix the plumbing, build the lists, then spend the money.
Remarketing is usually launched in one afternoon with a single all-visitors list and a default duration. It then runs for years. The order below costs an extra week at the start and changes what the channel is worth.
Before a single tag fires
- Preference signal traced from arrival through to the advertising tag
- The connection tested as a visitor, not assumed from a plugin description
- Restricted categories checked against the vertical you operate in
- Existing audiences reviewed for records whose origin cannot be established
Building the lists
- Audiences defined by meaningful actions rather than by having arrived
- Duration set to the decision window for that specific service
- Converters suppressed so the budget stays on undecided people
- Customer lists used only where the basis is recorded
Running it
- Frequency capped so the campaign persuades rather than irritates
- Creative that acknowledges the visit rather than restarting the pitch
- Results reconciled against work that reached the business
- Audiences reviewed on a cycle, not left running until the budget ends
Six things that happen before we spend anything
None of these produces an impressive slide. Together they are the difference between a channel you can explain to a customer who asks why they keep seeing your ads, and one you cannot.
The preference signal wired through
We trace the signal from the request header to the consent layer to the tag, and connect it where the chain is open. Then we test it the way a visitor would experience it.
Source: Global Privacy Control and Civil Code 1798.135
Audiences rebuilt from behavior
All-visitors lists get replaced with audiences defined by what somebody actually did — read a service page, started a form, called without booking.
Source: Google Ads Help — data segments
Duration matched to the decision
Membership length is set per service against how long that decision actually takes, rather than left on a default that keeps paying to reach people who moved on months ago.
Related: the demand timing question is covered on market research.
Converters suppressed
People who already booked come out of the audience. Continuing to advertise to a customer is not brand presence, it is a message that nobody is watching the account.
Related: the suppression list comes from your CRM, which is why the two systems have to agree.
Uploads checked before they move
Customer lists are checked against the platform's own terms and against what those people agreed to when you collected their details. Where the basis cannot be established, that portion does not go.
Placement provenance still applies
Remarketing runs on the same networks as prospecting display, so the same question applies: can anybody confirm where the ad ran? We check, and we exclude what cannot be verified.
Related: the full method is on display advertising.
We will make your audience smaller on purpose.
Audience size is the easiest number in the channel to grow and the least connected to whether it works. Removing people who never came close, people who already bought, and people who asked to be left alone produces a smaller list and a better one. We report the removals as the result, because they are.
We will also tell you when remarketing is the wrong spend. If the site is not getting enough qualified traffic to build a meaningful list, remarketing is a magnifying glass over an empty page. Fix the traffic first; the finding costs us the campaign and we report it anyway.
Remarketing sits alongside paid search, display and the rest of the digital program, under the same accounting discipline.
Every layer has a source, or it comes out.
A remarketing audience is not one thing. It is several sources stacked together, and the useful discipline is being able to name each one — because the layer nobody can account for is the layer that creates the problem.
| Audience source | How it is usually set up | Accounted for How we set it up |
|---|---|---|
| Site visitors | One list, everyone who loaded a page | Separate lists by what the person actually read and how far they got |
| Form abandoners | Not built at all | Built deliberately — the highest intent audience on most sites |
| Callers who did not book | Invisible to the ad platform | Reachable where your call and CRM records support it |
| Customer list uploads | Uploaded once, terms unrecorded | Basis established and recorded before the file moves |
| Existing converters | Left in, still being advertised to | Suppressed, so the budget stays on undecided people |
| Records of unknown origin | Inherited and never questioned | Removed — a record we cannot explain is one we will not target |
The last row is the one that shrinks the audience and the one worth doing anyway. A list you can describe line by line is a list you can defend to a customer who asks why they keep seeing you — and that conversation happens more often than most operators expect.
Four remarketing line items you can stop paying for.
One is a number that measures reach rather than intent. One is a setting nobody revisited. One is a practice the medium makes easy and nobody enjoys receiving. The fourth is a data transfer done without checking.
Audience size reported as progress. Growing a list is trivial — loosen the definition and it grows. A list of everyone who loaded a page is mostly people who were never close, and the size of it tells you about your traffic rather than about your remarketing.
Default membership duration. Left alone, it keeps paying to reach people whose need was resolved months ago. For a service with a short decision window that is the majority of the audience within weeks, and nobody notices because the campaign keeps delivering impressions.
Following people who asked you to stop. Sometimes deliberate, far more often an unconnected chain nobody tested. Either way the person experiences the same thing, and in California the request they made has statutory standing. This is the one item on the list that is a risk as well as a waste.
Uploading a customer list without checking the basis. The file moves in a few clicks and the question of what those people agreed to lives in your records, not the platform's. Doing it in the wrong order is how a routine task becomes a problem nobody planned for.
The pattern beneath all four: the channel makes it easy to reach more people, when the discipline is deciding who to stop reaching.
- California Civil Code section 1798.135 — opt-out methods and preference signals
- California Civil Code section 1798.120 — right to opt out of sale or sharing
- California Attorney General — California Consumer Privacy Act
- California Attorney General — CCPA regulations
- Global Privacy Control
- IAB Tech Lab — Global Privacy Platform
- Google — Restricted targeting in personalized advertising
- Google Ads Help — About your data segments
- Google Ads Help — Customer Match policy
- Google Ads Help — Google Ads policies
- Google Ads Help — Exclude specific webpages and videos
- FTC — Privacy and security business guidance
- FTC — Advertising and marketing business guidance
- IAB Tech Lab — ads.txt, Authorized Digital Sellers
Remarketing, answered against the published text
Every answer below links to the statute, guidance or platform policy it rests on, so you can check it without taking our word for it.
What is an opt-out preference signal, in plain terms?
It is a setting a visitor turns on once in their browser, after which the browser tells every site they visit that they do not want their personal information sold or shared. Global Privacy Control is the best known implementation. California's Civil Code section 1798.135 treats such a signal as a legitimate way for a business to let consumers exercise that right, as an alternative to carrying the "Do Not Sell or Share My Personal Information" link. So it is not a courtesy — it is a recognized mechanism. The equivalent question for your own records is on our CRM page.
Our consent banner handles this. Doesn't that cover us?
It covers receiving the signal. Whether anything acts on it is a separate question, and it is where the chain usually breaks. Consent layers are installed to manage banners and cookie categories; advertising tags are often installed years earlier by someone else, and fire on page load without consulting the consent state. Both components work correctly in isolation and no error is produced. Test it as a visitor would — enable the signal, load the site, watch whether the tag fires. The Attorney General's CCPA guidance is the reference. Same shape of silent failure as on website maintenance.
We are not in California. Does any of this apply to us?
Possibly, and it depends on where your visitors are rather than only on where you are. State privacy law has expanded well beyond California and continues to move, and several states recognize universal opt-out mechanisms in some form. We check the states you actually serve rather than assuming a single national rule, and we generally treat the strictest applicable position as the working default — maintaining several different behaviors is how inconsistency creeps in. The California regulations are the most developed reference. This is not legal advice; your counsel should confirm your position. Territory research is covered on market research.
Can we upload our customer list to build an audience?
Often yes, and the order matters. Google's Customer Match policy governs what may be uploaded and on what basis, and the harder question sits in your own records: what did these people agree to when you collected their details, and does that cover advertising to them on a third-party platform? Establish that before the file moves, record it alongside the audience, and where it cannot be established for part of the list, that part does not go. The consent record itself lives in your CRM.
Why do our ads follow people around for months?
Because membership duration was left on its default and nobody revisited it. Google documents how data segments work and how long a person stays in one, and the default is far longer than most home services decisions take. Somebody who needed an emergency repair in March has either solved it or moved on; paying to reach them in June is not persistence. Set duration per service against the real decision window, and cap frequency so the campaign persuades rather than irritates. Timing is a demand question, covered on market research.
Are there people we are not allowed to target?
There are categories you may not build audiences around. Google's restricted targeting policy limits or prohibits personalized advertising in defined sensitive areas, on the basis that targeting around them is intrusive regardless of intent. This catches medical and financial verticals more often than operators expect, and it is far cheaper to establish before an audience strategy is designed than after a disapproval. Design around the restriction rather than testing where the edge sits — the same posture we take on policy in paid search.
Should we advertise to people who already bought?
Almost never with a remarketing budget, and the exception is a genuine second purchase you can name. Advertising to an existing customer asking them to become a customer signals that nobody is watching the account, and it spends money that belongs on people who have not decided. The suppression list comes out of your own records, which means the ad platform and your CRM have to agree with each other. Google's data segment documentation covers the mechanics of exclusion.
Does remarketing have the same placement problem as display?
Yes — it runs on the same networks, so the same question applies: can anybody confirm where the ad actually ran? The IAB Tech Lab's ads.txt standard exists because counterfeit inventory and domain misrepresentation are real enough to warrant a global specification, and the check is free. Audience precision does not protect you from an unverifiable placement; a well-built list shown on inventory nobody can vouch for is still money spent on an unexamined assumption. The full method is on our display advertising page.
How do we know remarketing is actually working?
By reconciling what the platform reports against what reached the business, and by being honest about attribution. Remarketing sits late in a journey, so it is credited generously by last-click and platform-side models — a person who was already returning gets counted as a conversion the channel produced. Compare reported conversions against booked work, hold a portion of the audience out where volume allows, and watch total booked work rather than platform conversions alone. The reconciliation discipline is set out on paid search, and Google's own conversion tracking documentation covers what the platform is and is not counting.
Is remarketing worth it for a local service business?
Usually yes, once there is enough qualified traffic to build a meaningful list, and it is one of the better-value channels when the plumbing is right. It is the wrong first purchase when the site is not yet receiving enough of the right visitors, because remarketing can only work with people who already arrived — it magnifies what exists rather than creating it. It is also wrong where the service is a one-time emergency purchase with no repeat window. General expectations for handling personal information are covered in FTC privacy guidance. Start with an honest read of your traffic, which is what the A.R.C. Report is for.
Find out who is in your audience and why.
The A.R.C. Report covers your whole marketing position, and where remarketing is in the plan we look at how the audiences were built, whether the preference signal reaches the tag, and how much of the list is people who were never close or already bought. Findings are yours whether or not we work together.
- Preference signal traced from the browser through to the advertising tag
- Every audience checked for records whose origin cannot be established
- Membership duration compared against the real decision window
- Converters checked for suppression against your own records
- Customer list uploads reviewed for a recorded basis
- A straight answer if remarketing is the wrong spend for you right now
Explore the wider program: AI SEO, all services, paid search, display advertising, landing pages and the A.R.C. Report.
Tell us the site and the markets you serve, and we will tell you what your remarketing is currently buying.
No cost, no commitment. We will follow up by email or phone to walk you through the findings.
