AI Is Powerful, But Expertise Still Matters

3rd Oct, 2026

AI has created a fascinating new phenomenon in digital marketing.

Suddenly, everybody is an expert.

A business owner puts their website into ChatGPT or Claude, asks it for an SEO audit, gets back 47 recommendations and forwards the whole thing to their agency with some variation of:

“Why haven’t you done all of this?”

Or they upload a Google Ads report and ask AI what’s wrong.

Or they ask ChatGPT which keywords they should target.

Or they generate a landing page.

Or a content strategy.

Or a Meta campaign structure.

And somewhere between pressing Enter and reading the beautifully formatted response, they become convinced that their agency apparently has no idea what it’s doing.

So, as an agency that actively uses AI every single day, perhaps we should clear something up.

AI is incredibly useful.

It is also not an expert at everything.

And having access to AI does not suddenly give you 10, 15 or 20 years of experience in digital marketing.

We use AI too. Probably more than you do.

This is not an anti-AI rant.

Quite the opposite.

We use AI extensively.

We use it to interrogate data, analyse patterns, structure information, research, brainstorm, troubleshoot, summarise, compare, build frameworks and accelerate work that previously took significantly longer.

It makes experienced people faster.

What it does not do is magically replace the experience required to know whether its answer is actually correct, relevant, prioritised properly or even applicable to your business.

That distinction matters.

When I audit a website, I don't simply paste the URL into ChatGPT and ask:

“What is wrong with this website?”

An actual SEO investigation may involve information from:

Google Search Console
Google Analytics
SEMrush
Search Atlas
Screaming Frog
Meta SEO Inspector
SEOQuake
Google Ads
Competitor research
Technical crawls
Page indexing data
Conversion data
Existing strategy documents
Historical performance
Internal SOPs
Previous changes made to the website
And, importantly, years of experience interpreting all of the above.

Then AI can help us interrogate that information.

That is very different from asking an AI model to look at a website from the outside and assuming its response constitutes a professional audit.

AI might get you 80–95% of the way there in some situations.

The problem is that you often need expertise to identify the other 5–20%.

And sometimes that 5% is the difference between fixing the problem and making it worse.

AI can give you an answer. That doesn't mean it's the right answer.

Let's use a medical example...

You could spend six months putting symptoms into ChatGPT.

You could tell it about your headaches, fatigue, blood pressure, sleep, diet and whatever else you're experiencing.

Eventually, you might have a very intelligent conversation about several possible causes.

That's useful.

But ChatGPT can't reach through your laptop and draw blood.

It can't physically examine you.

It can't run the tests required to confirm which theory is correct.

You still need the diagnostic tools and a qualified person capable of interpreting the results.

Digital marketing works the same way.

AI might tell you:

“People are probably searching for ‘digital health audits’.”

Cool.

Now let's pull actual search-volume data.

Let's look at Google Search Console.

Let's look at Semrush.

Let's look at Search Atlas.

Let's look at related queries.

Let's look at intent.

Let's look at competition.

Let's see whether your website has the authority to rank for it.

Let's see whether those searches actually have commercial value.

And sometimes, after doing all of that, we discover that AI's perfectly plausible recommendation is wrong.

Not because AI is useless.

Because plausible is not the same thing as proven.

Your car probably has a battery problem. Probably.

The same thing happens when something goes wrong with your car.

The car won't start.

It looks like the battery.

AI might tell you it's probably the battery.

Your neighbour might agree.

You might even replace the battery.

Except the real problem could be the alternator.

Or a fuse.

Or a connection.

Or something else entirely.

A good mechanic combines symptoms, diagnostic equipment, technical knowledge and experience to determine what is actually wrong.

Marketing is no different.

Tools + data + experience + context = diagnosis.

AI is one part of that equation.

Not the entire equation.

“But AI says…”

This phrase is becoming the digital marketing equivalent of nails on a chalkboard.

Because the immediate question should always be:

What information did you give it?

An AI model can only reason from the context and information available to it.

If you give it incomplete information, it can produce an extremely convincing answer based on incomplete information.

That does not make the answer useless.

It does mean you need to understand its limitations.

Ask AI to review your Google Ads account without knowing your margins, lead quality, sales cycle, attribution model, offline conversions, business objectives, geographic priorities or historical performance and it can still happily produce recommendations.

Some might be good.

Some might be terrible.

And without experience, you may not know which is which.

Yes, you can run your own marketing.

Here's the uncomfortable part for agencies:

You absolutely can do a lot of this yourself.

You can build a Google Ads campaign yourself.

You can launch Meta Ads yourself.

You can write content with ChatGPT.

You can build landing pages.

You can research keywords.

You can change title tags.

You can install plugins.

You can modify your website.

You can analyse reports.

There has never been more information available to business owners who want to learn digital marketing.

But access to the tools has never been the thing you were paying an agency for.

You're paying for knowing what to do with them.

You can buy the same drill a professional carpenter owns.

That does not mean the cabinet you build this weekend is going to look like theirs.

You are not paying me for five minutes.

There's a famous concept that gets repeated across practically every skilled profession:

You are not paying someone for the five minutes it took them to do the job. You're paying them for the 15 years it took them to learn how to do it in five minutes.

Digital marketing is no different.

Sometimes I can look at an account and identify a problem almost immediately.

That doesn't mean the problem was easy.

It means I've probably seen a variation of it dozens of times before.

Experience compresses time.

That's what expertise does.

The same applies to your accountant.

Your lawyer.

Your doctor.

Your engineer.

Your mechanic.

Your IT provider.

And, presumably, whatever professional service your own business sells.

Which brings us to something worth considering.

Imagine one of your customers putting a question about your field into ChatGPT, forwarding you the response and then explaining how you should be doing your job.

You'd probably not have a very good response to that…

Your AI audit is not automatically our new scope of work.

There is another issue that AI has made significantly worse.

The never-ending audit.

A client runs their website through an AI tool.

Then another AI tool.

Then another SEO tool.

Then their cousin runs something.

Then someone on LinkedIn posts an SEO checklist.

Suddenly the agency receives an email containing 20 new “urgent” tasks.

Here's the problem:

Your agency agreement still exists.

If you are paying your agency for one hour of SEO work per month and you send them 20 hours of new SEO work, those 20 hours don't magically fit into the one hour you're paying for.

You have three choices:

Prioritise the work and complete it progressively within the existing retainer.
Approve additional budget.
Increase the scope of the agreement.

What you cannot reasonably expect is unlimited output from limited contracted time.

If your agreement gives you a certain number of hours, deliverables or projects per month, that is what you're buying.

Not every recommendation any AI platform produces between now and the end of time.

An agency is not your in-house employee.

This misunderstanding causes enormous frustration on both sides.

An agency and an internal marketing department operate differently.

If you want somebody available throughout the day who you can message at 10:12 with a request and expect to have it implemented by lunchtime, what you're describing is usually an employee.

Hire one.

But understand the economics.

A genuinely experienced SEO specialist, PPC specialist, strategist, developer, designer or paid-social specialist represents a significant monthly salary, and one person is still unlikely to be excellent at every discipline.

With an integrated agency, your monthly investment may give you access to:

SEO specialists
Google Ads specialists
Meta specialists
Strategists
Copywriters
Designers
Developers
Analytics expertise
Account management
Conversion optimisation
Marketing technology expertise
Tools & Systems (that are usually charged in dollars or pounds)

You are sharing those specialists' time with other clients.

That's part of why the model is economically viable.

So yes: when you send an agency 20 new tasks, they need to be prioritised against the scope and hours available.

That isn't incompetence.

That's maths.

And please stop randomly changing things because ChatGPT told you to.

This one is important.

If you have access to your website, ad accounts, Google Business Profile, tracking, Merchant Center, analytics setup or any other marketing infrastructure, please understand that changes have consequences.

Do not read an AI recommendation at 11pm, log into WordPress and start ripping things apart because ChatGPT said your H1 needs changing.

Do not restructure campaigns without telling the people managing them.
Do not install random plugins.
Do not rewrite high-ranking pages.
Do not take out content because “it looks like a lot and I don’t want to read that much”.
Do not change conversion tracking.
Do not remove scripts.
Do not modify URLs.
Do not “fix” technical SEO issues without understanding dependencies.

Because then, three weeks later, when traffic drops, conversions disappear, or tracking stops working, everybody gets to spend hours figuring out what actually happened.

If you're working with an agency, work WITH the agency.

Send them the AI recommendation.

Ask:

“Does this make sense?”

“Can you please have a look at this and give me feedback”

Good agencies should be perfectly comfortable answering that question.

Sometimes the answer will be:

“Yes. That's a great recommendation. We should implement it.”

Sometimes:

“Yes, but it's low priority compared with these three things.”

Sometimes:

“That recommendation is technically correct, but inappropriate for your situation.”

And sometimes:

“No. Here's the data showing why.”

That conversation is valuable.

Blind implementation isn't.

The best agencies aren't fighting AI. They're using it.

AI is not going away.

Nor should it.

It is one of the most powerful productivity tools we've had access to in years.

It makes research faster.
It makes analysis faster.
It makes ideation faster.
It helps us interrogate huge volumes of information.
It can challenge assumptions.
It can help experienced professionals consider possibilities they may have missed.
And it is improving unbelievably quickly.

But here's the bit everyone seems desperate to skip:

AI works best when the person using it knows what they're doing.

A digital marketing specialist using AI isn't equivalent to somebody with no marketing experience using AI.

Just as a doctor using an AI diagnostic assistant isn't equivalent to you putting your symptoms into ChatGPT.

The technology may be the same.

The context, judgement and expertise surrounding it are not.

Challenge your agency. Just don't confuse access to AI with expertise.

Clients should absolutely challenge their agencies.

Ask questions.
Ask what the priorities are.
Ask whether something AI suggested is worth investigating.
Ask for data.
Ask for explanations you actually understand.

Hold your agency accountable for what they're contracted to deliver.

A good agency shouldn't be threatened by that.

But there is a difference between:

“ChatGPT identified this. Can we investigate whether it's relevant?”

and:

“ChatGPT says you're doing this wrong. Fix these 37 things immediately.”

One is collaboration.

The other is somebody using a general-purpose AI tool to tell a specialist how to do the job they hired them to do.

So yes.

Use ChatGPT.

Use Claude.

Run the audits.

Ask questions.

Learn.

Become a better-informed client.

We genuinely love informed clients.

But when your AI tool spits out 50 recommendations, don't assume you've suddenly uncovered 50 examples of your agency being incompetent.

Sometimes you've uncovered great opportunities.
Sometimes you've uncovered low-priority work.
Sometimes you've uncovered things we've already evaluated.
Sometimes the recommendation isn't relevant.
Sometimes it's flat-out wrong.
And sometimes the thing that looks like a five-minute fix is sitting behind ten years of experience, five different data sources and a specialist who knows exactly which lever to pull.

AI gives you information.

Expertise tells you what to do with it.

There is still a difference.

And for now, it's a pretty important one.

<< Back to News Index