Optimizing Google and Meta ads with AI: how it gets concrete

Anyone who wants to optimize Google and Meta ads with AI today is offered two very different things: the automation of the advertising platforms themselves – and tools that bring your own company knowledge into the campaigns. Only the second is still a competitive advantage. After all, everyone now uses Smart Bidding and automated placements; they reliably optimize for the goal you give them, but know nothing about your margins, your ideal customers or your capacity. That is exactly where the interesting work begins.

Why “AI optimization” in the platforms isn't what you mean

The bidding automation of the ad systems is good. It is better at hitting a click target than any person with a spreadsheet. But it only knows what happens in its own account: clicks, conversions, signals from your website. What it doesn't know: that product A brings in three times the contribution margin of product B, that inquiries from a particular region regularly don't lead to a deal, that your workshop is fully booked for the next six weeks.

That is why we consider the phrase “the AI optimizes our campaigns” a dangerous oversimplification. The automation doesn't optimize your business, it optimizes your target. If the target is “as many form submissions as possible”, that is exactly what you get – even if half of them are useless.

Which company knowledge the campaigns actually lack

• Contribution margins per product or service – not revenue, but what is left over
• Closing rates per source – which campaign brings inquiries that also turn into orders
• Exclusion criteria – industries, regions and inquiry types you don't want to serve at all
• Capacity and season – when additional inquiries create value and when they only create work
• Your customers' language – how inquiries, support tickets and offers are actually worded

This knowledge exists in every company. It just rarely exists in a form that can be translated into campaigns. That is precisely why good ads work today depends on a question that at first glance has nothing to do with advertising: how well is your company knowledge structured?

How an AI assistant with company knowledge closes the gap

Our approach connects an AI assistant to the company's structured knowledge base – as described in the article Obsidian as a second brain – and to the data from the CRM and the offer system. Four applications that hold up in practice:

1. Reading search term reports that nobody else reads

Every Google Ads account contains search queries you would rather not pay for. Finding them is not an intellectual achievement but diligent legwork – and that is exactly what an assistant that knows your exclusion criteria is good at. It flags candidates, a person decides. This is the least spectacular and at the same time most reliable lever on the cost per inquiry.

2. Ad copy from real customer wording

Generic AI copy sounds like AI. Copy fed by your own offers, support replies and inquiries sounds like you – and addresses the objections your customers actually have. The difference is not in the model, but in the material.

3. Feeding back lead quality instead of lead quantity

The most effective intervention of all: report back to the ad platform not the form submission, but the qualified lead or the order. Then the automation optimizes for deals instead of clicks. This requires a clean connection between website, CRM and ad account – integration work, not a marketing trick, and in our view the point at which AI optimization starts to make sense at all.

4. Creatives in series, not as one-offs

On Meta, creative variety matters more than fine-tuning audiences. Anyone who has only two motifs in circulation wears out their audience. For production in series we use our own platform AI Ad Studio – graphics and videos in line with the brand, instead of commissioning each motif individually.

What we consider marketing – and not substance

• “Fully automated AI campaign optimization.” There is no automation that decides which customer is valuable without your business knowledge. Anyone who promises this is selling the platform's automation as their own work.
• Percentage promises without a baseline. A double-digit percentage reduction in click costs means nothing without stating the starting point, the period and the lead quality.
• Daily optimization on thin data. Anyone with a low double-digit number of conversions per month mainly measures chance when adjusting daily. Restraint is the better optimization here.
• Scripts that shift budgets automatically. Without a significance check they create movement, not results.

Google or Meta – where does the lever lie in each case?

The two channels reward different work, and this is regularly confused. On Google, what matters is how precisely you delimit demand: which search queries do you want to serve, which ones definitely not, and does the landing page match the query? Here the gain lies in leaving things out – in consistently maintained exclusions and in landing pages that answer exactly the question that was asked.

On Meta there is no demand you could tap – you create it. That is why the motif decides here, not the audience setting. The systems now reliably find your audience themselves; what they need is enough material to learn from and a clean signal of what counted as a success. Anyone on Meta who works on audiences instead of motifs is working on the wrong end.

The same principle applies to both: the AI delivers drafts and anomalies, a person makes the decision. An assistant that suggests twenty ad variants is useful. A script that runs them without review is a budget risk.

A process as it actually looks with us

So that this doesn't stay abstract, here is a typical weekly rhythm for the copy work: the assistant pulls the new search queries and the inquiries from the form for the last seven days. It checks them against the stored exclusion criteria and presents two lists – terms that should be excluded, and phrases that come up conspicuously often and don't yet appear in any ad.

The second list gives rise to copy variants whose building blocks come from real offers and support replies. A person reviews them, deletes half and publishes the rest. The whole process takes just under an hour a week instead of half a day – and the suggestions are better because they are based on material the ad platform doesn't even know.

A realistic build-up in 30 to 60 days

• Week 1: check the measurement. What is counted as a conversion, what is counted twice, what not at all?
• Week 2: define quality. What is a good lead – and who decides?
• Weeks 3–4: build the feedback channel: qualified leads and orders back to the ad platforms
• Weeks 4–6: connect the assistant to the knowledge base and CRM, set up search term and copy work
• From week 6: creative production in series, a fixed monthly evaluation by contribution margin

How you should measure success

Not by the cost per click. Not by the number of inquiries either. The two key figures that count are the cost per qualified lead and the contribution margin per euro spent. Both require sales data to flow back – which is why ads optimization in our projects is almost always an integration topic first and only then an advertising topic. How we build such connections is described under services, and you can find examples under case studies & apps.

Conclusion

The platform AI is a tool that is equally available to everyone. Your edge comes from what only you have: knowledge of margins, customers and deals. Those who structure this knowledge and feed it back into the campaigns are truly optimizing. Those who just switch on the automation and hope for the best are optimizing their competitors' results at the same time.

Frequently asked questions

Does AI replace manual campaign management?

No. It shifts the work from routine tasks towards decisions: goal setting, defining quality, budget allocation and creative strategy. The effort doesn't fall to zero, it is just used more valuably.

From what advertising budget is this effort worthwhile?

The feedback channel for lead quality is almost always worthwhile, because it also benefits sales. Elaborate automation and creative production in series usually only pay off from a regular monthly media budget in the four-figure range.

What about data protection when CRM data flows back to ad platforms?

That is a central point and has to be solved properly. As a rule, no plain-text data is transferred, but pseudonymized identifiers or pure event notifications. We clarify the scope and legal basis together with you before implementation.