How can you tell if an agency really understands AI visibility?

On Monday, they couldn’t explain why your traffic had dropped. By Thursday, they understood how every major language model chooses which brands to recommend.

You’d think there’d be a course or something in between.

But no. Just a new page on the website and a photo of them looking like they know something.

I admire the confidence…

To judge an agency’s AI visibility expertise, ask what experience sits behind the proposal, what it has studied and tested, and what it still doesn’t know.

Why are so many agencies suddenly AI visibility experts?

Because clients have started asking, and the demand is real. Pew Research Center found that Google users clicked a traditional result in 8% of visits when an AI summary appeared, against 15% when one didn’t. Search behaviour is shifting, and agencies are responding.

The question is how much real understanding sits behind the response. Or is it just this guy?

The Pew analysis tracked the March 2025 browsing of 900 US adults, and it found that people clicked a link inside the AI summary itself in only 1% of visits. Google has disputed the findings, and this is US data from a single month, so treat it as a signal rather than a forecast. It does explain why clients are asking questions.

Now, I work in this industry. Friday offers AI visibility services. You would be entitled to ask whether we’ve done the same thing. It’s a fair question, so I’ll answer it.

What experience should sit behind an AI visibility proposal?

A track record you can date, training that goes beyond reading about the subject, and practical testing on real client sites. Credentials give context, but they do not replace explaining a recommendation or showing that it works. Before signing, business owners and marketing directors should ask what the agency has studied and tested, and what it still doesn’t know.

In my case, the interest predates the current rush by several years. I have spent more than two decades working across digital marketing, including SEO, paid media, analytics and strategy. Over that time, I have watched plenty of developments arrive carrying predictions of the death of everything that came before them.

Experience teaches you to investigate before ordering the funeral flowers.

My own interest in AI led me to undertake training and qualifications in AI and machine learning, alongside cyberpsychology. I wanted a better understanding of the technology and of the people using it. Both matter when your job involves helping businesses reach customers.

I have also written two books: Hiding Dead Bodies: Your No-Nonsense Path to SEO Glory, and Driving Blindfolded: No-Nonsense Digital Marketing Strategy for the AI Age. The titles are slightly more dramatic than my average working day, which involves more spreadsheets, but are very much in line with my no-bullsh*t approach to all of this.

Those qualifications and books give context to my approach. They do not excuse me from explaining a recommendation or demonstrating whether it works. A certificate cannot do your client’s thinking for you.

What does five years of preparation actually look like?

It looks like dated, public work and a steady alignment of SEO practice with where search was heading. Clearer content, accessible websites, sound technical foundations and evidence of expertise.

At Friday, that record starts with an article in July 2021 on AI changing digital marketing. Anyone can check it, which is the point.

That article, published on Friday’s website in July 2021, covered machine learning, semantic search, content creation and the implications for marketers. It gives our involvement a date.

It would be stretching things to claim we knew exactly how today’s AI search products would work. We could see that machines were becoming more capable of interpreting language and that this would affect how information was discovered.

Our preparation meant aligning our SEO approach with that direction. These were sensible investments before anyone started asking how to appear in ChatGPT.

In January 2025, I wrote specifically about how accessibility standards could make websites easier for AI systems to interpret. The connection between useful information, readable structure and machine understanding has been part of our thinking for years.

We have continued adapting as the platforms have developed. Preparation gives you somewhere useful to start; it does not mean you can stop learning.

Is SEO still part of AI visibility?

Yes. And don’t listen to anyone who says it’s not. Google’s guidance says established SEO best practices remain relevant for AI Overviews and AI Mode, and that no specific optimisation is required for either. A page still needs to be indexed and eligible to appear with a snippet. That applies to Google’s products, so it shouldn’t be stretched to every AI platform.

I become furious when someone explains AI visibility by declaring SEO irrelevant. Technical accessibility, useful content and clear information have not suddenly become old-fashioned.

If your website fails to explain what you sell, who you serve or why someone should trust you, there is work to do. Renaming that work will not complete it.

For example, a software business might describe its product as an innovative solution that empowers organisations. A prospective buyer needs to know whether it handles staff scheduling, integrates with payroll and supports their particular sector.

I would start by answering those questions properly. The machine can wait while we establish what the business actually does.

For a business comparing proposals, I would look for a specific diagnosis. Perhaps service information is missing, or the company is described inconsistently across sources. Recommendations should address those findings. Ask what needs fixing before approving another six months of articles nobody particularly asked for.

What is genuinely new about AI visibility?

Generated answers raise questions that a ranking report cannot answer: whether your business is mentioned, whether it is described accurately, which sources sit beside it, and whether the answer changes when the question gets more specific. Anyone selling AI visibility as exactly the same service under a new invoice heading is skipping that work.

Those are things I want investigated. An agency should also distinguish between information learned during model training and information retrieved when a user asks a question. Updating a page does not mean retraining a language model.

Nor does being cited mean a platform has endorsed everything your business does.

There is some research on what influences citations. A KDD 2024 study by Aggarwal and colleagues found that adding citations, quotations and statistics to content raised its visibility in generative engine responses by up to 40% in their tests, while keyword stuffing did not help. It is one academic benchmark rather than a guarantee for any particular site, which is why testing in a client’s own market matters.

My training helps me interrogate those distinctions. Practical testing helps establish what they mean for a particular client. Neither gives me access to a secret switch marked Recommend This Company.

I would have mentioned it. And my books would have been shorter.

How should an agency measure AI visibility?

It should show you the questions it tested, why they represent your customers, which platforms it used and what changed between repeated runs. One favourable answer proves little. A visibility score means nothing until someone explains what sits underneath it, so ask for a record of prompts, dates, platforms and sources.

If I were choosing an agency, I would ask it to walk me through an assessment. Which questions represent my customers? Why were those questions selected? Which platforms were tested? Were the tests repeated, and what changed between runs?

One favourable answer is encouraging. It is a poor foundation for a twelve-month commercial commitment. I would want a clear account of the limitations alongside the record.

The reporting should distinguish a brand mention from a citation, a referral visit and an enquiry. Each tells you something different. None should quietly become revenue halfway through a presentation.

At Friday, the commercial question matters to me… are we helping suitable customers discover the business and make an informed decision? Visibility deserves attention, but the client still needs customers.

What should you ask an agency before you appoint it?

Ask for relevant experience, an explanation of the proposed work, the evidence behind it and an honest account of what remains uncertain. A credible practitioner can discuss all four without looking as though you have caught them out. Then ask what they have changed their mind about and what they tested that failed.

• What experience, study and testing sit behind this proposal?
• What specific problem did you find, and which recommendation addresses it?
• Which questions, platforms and dates did the assessment cover, and was it repeated?
• How will reporting separate mentions, citations, referral visits and enquiries?

I have no objection to agencies developing new services. They should. Clients need partners who keep learning as their markets change.

What concerns me is certainty arriving before the evidence.

Ask what they have changed their mind about. Ask what they tested that failed. Those answers can tell you more than another confident prediction.

That is the standard I want applied to my own work and to Friday’s. Our experience, training and preparation matter because they help us make better decisions for clients. They have to earn their place in the work.

Before you appoint an AI visibility agency, have that conversation. You should leave understanding what needs attention, how progress will be assessed and why the people proposing it are equipped to help.

The website is the easy part to update.

Weighing up some SEO/GEO agencies? Talk to Friday about what needs attention first.