How Exhibit A Media works as an AI marketing agency with two humans in the loop

An AI agent wrote the first draft of this article. Two more agents checked it, and Frank Alfano read it and approved it before it went live.

Exhibit A Media is an AI marketing agency run by two people, Frank Alfano and Jana Smith. In Frank's words, we are humans working with AI, not entirely AI. If you run a law office, a paralegal practice, a dental clinic or an accounting firm, and you are wondering whether an agency that uses AI is doing better work or cutting corners, here is exactly how we split the work between AI and people. You will also see why that split gets you better results than either would alone, and what to ask any agency about its own use of AI, whether or not you ever hire us.

What "human in the loop" means here

Human in the loop means a person sits inside an automated process at the points where decisions get made, and the process can't move past those points until that person says yes. The machine does the work between the checkpoints. The person owns the checkpoints.

At Exhibit A Media, the work between the checkpoints is research, drafting, design and checking. AI agents do most of it. The checkpoints belong to Frank and Jana.

Frank sets the strategy and direction for each client, and he approves what ships. Nothing is published, sent to a client or spent on ads without his approval. Jana gets to know each client's business: what you do, who your clients are and what you need the website and marketing to accomplish. She reviews the work against that, and she makes sure the project delivers what you came for.

So the loop has two people in it, and they are the same two people you talk to. You won't find an account manager relaying messages to a production floor you never see. The agents recommend and the people decide, and that order doesn't flip.

An agent can write a perfect paragraph, and it still doesn't reach your website until Frank approves it. He reads it first.

Who does the work, human and otherwise

The two people

Frank Alfano is the founder and chief developer. He spent 33 years in Ontario courtrooms (1990 to 2024), holds an LL.B. from the University of London and an LL.M. from Osgoode Hall Law School, and as chief developer he is responsible for the websites and systems we build. His phone number and email are on his team page.

Jana Smith is a partner and head of business development. She is a Law Society of Ontario licensed paralegal and teaches continuing professional development courses for legal professionals through Bootcamp CPD. Her background also includes running a retail store and leading business development for a web design firm. When a lawyer or paralegal describes how their practice runs, she knows the work from their side of the desk.

The agents

Our AI agents have names, borrowed with affection from fictional characters and a few famous real people. That makes the work log easier to read: "Ernest sent Ken's draft back" tells us more at a glance than "review step four returned a failure." It also means a review desk where Hemingway tells the writer his sentences are too long, which is about as close to typecasting as software gets.

Here are the agents most likely to touch your work, by what they do.

Agent	What it does
Dr. Faye Miller	Consumer research: what your customers are searching for, and the evidence behind it
Tony Stark	Keyword research: which searches each page should target, with volume and difficulty
Bugs Bunny	Competitor research: who you are up against and where you can win
Mark Zuckerberg	Trend research: what is travelling on social platforms this week
Don Draper	Head of creative: writes the brief and picks the angle
Ken Cosgrove	Staff writer: drafts long-form articles from the brief
Ernest Hemingway	Article review: readability, structure and whether each claim has a source
Fog Horn Leg Horn	SEO engineer: keywords, metadata, internal links and technical checks
Salvatore Romano	Art director: featured images and social graphics
Peggy Olson	Copywriter: turns an article into social posts
Adam Mabee	Builds websites
Luther Stickel	Independently reviews each build before it is released
Kevin Mitnick	Website security checks
Michael Ginsberg	Quality audits of the sites we run

Four rules hold the roster together. The agent that writes something never reviews its own work. Reviewers name the fixes and leave the rewriting to the writer. Agents recommend, and Frank and Jana decide. And no agent publishes, spends money or emails a client on its own.

That last rule matters most to you. The agents are fast and tireless, and none of them has the authority to commit you to anything.

How this article moved through the chain

This article went through every step below before you read it.

1. Keyword research came first: which searches a business owner types when they wonder about agencies and AI, and how often those searches happen in Canada.
2. Don Draper wrote the brief. It named the reader, the one promise the article had to keep, the facts the writer could state and a list of claims the writer was not allowed to make, including any speed multiple or results figure we couldn't prove.
3. Ken Cosgrove wrote the draft, starting with three candidate opening lines, the strongest of which opens this article. Each outside source was opened and read before Ken used it, and claims Ken couldn't confirm on the page were cut.
4. Ernest Hemingway reviewed the draft for readability, structure and sourcing. He checked each factual claim against the page it cites and named any fixes he wanted.
5. Fog Horn Leg Horn ran a separate SEO pass on the same draft: the search intent, the keywords and where they sit, the page title and description, and whether each internal link points somewhere useful.
6. Ken made the fixes the reviewers named, and only those.
7. Salvatore Romano made the featured image.
8. Frank read the finished article and approved it. Until he did, nothing could publish it.

Why the writer never grades its own paper

Separation of duties is the whole design. A writer reading its own draft sees what it meant to say, which is a poor way to find what it actually said. Ernest reads it cold. Fog Horn reads it for a different job entirely. Neither of them edits the text, so when a fix comes back, the record shows who asked for it and why.

Lawyers and paralegals will recognise the idea. It is the same reason someone other than the drafter checks a document before it is filed. The person who wrote it is the person most likely to read straight past the mistake.

We apply the same separation to websites. Adam Mabee builds, Luther Stickel reviews the build independently, and the release waits until the review passes. If you are vetting a developer, our article with questions to ask a web developer before you hire one is a good start.

What an AI marketing agency changes for your results

The honest benefit of an AI marketing agency is time, and the question that matters is what the agency does with the time it gets back. AI lets us do in hours what used to take days. We put those hours back into your work.

Here is where they go.

• More research before anyone writes. Consumer, keyword and competitor research happen first, on each piece, rather than being trimmed when a deadline gets close.
• More drafts to choose from. This article's opening was picked from three candidates. A writer working alone against the clock usually writes one and keeps it.
• More checks. Each draft gets an editorial review and a separate SEO review, and each factual claim is checked against its source before Frank sees it.
• More revisions, sooner. When a review names a fix, the writer makes it before the work goes to Frank, so the version he reads has already been through that round.
• Faster turnaround when something on your site needs attention, because the checking that used to eat a morning is already done when Frank sits down to decide.

Generic work is cheap now

The result is better work, and faster work as well. Speed alone would be a poor reason to use AI, because the fast, generic version of marketing has become nearly free. Anyone can ask a chatbot for an 800-word article on why a law firm needs a website and get one in the time it takes to type the request. Every other firm in your city can ask the same chatbot the same question, and they will all get roughly the same article.

When generic work costs almost nothing to produce, it is worth almost nothing to publish. The edge goes to whoever uses the time AI frees up to do the parts a chatbot can't do on its own: research that is specific to your practice, an angle that suits your clients, sources a reader can check and a person who knows your business deciding what goes out under your name.

We are on the leading edge of this, and the concrete version of that claim is the roster above: agents with one job each, none of which can approve its own work, and two people who sign off on the result. That structure is the advantage we sell, and it is what you get. If your interest is whether AI tools are naming your firm at all, our article on how to get ChatGPT to recommend your business by name takes that question on, and our free AI visibility check tests it.

What AI should never do on its own

Some jobs stay with people, at our shop and, we would argue, at any agency you hire.

Judgment calls stay with people. Which keyword to give up on because a client can't win it this year, whether an angle suits a criminal defence practice or a family practice, when to tell a client that the thing they asked for won't help them. An agent can lay out the options and the evidence. Frank and Jana make the call.

Approval stays with people. So do client relationships. You talk to Frank or Jana. An AI agent may draft a reply to your email, and nothing is sent until Frank approves it. Spending money stays with people too: no agent books an ad, buys a tool or commits your budget.

Why legal clients care most

Anything legal or regulated gets a person's eyes before it goes anywhere. Lawyers and paralegals answer to their regulator for what their marketing says. A language model can write a confident sentence about a credential, a result or a fee that isn't true. If that sentence lands on your website, you are the one who answers for it, whoever wrote the draft.

That is why each factual claim in our work needs a source a person can open, and why the brief for this article listed the claims we were not allowed to make. In Ontario, for example, the Law Society sets conditions on how lawyers and paralegals advertise fees, which we set out in why we publish our prices. This article isn't legal advice, and your own regulator's rules are the ones to read.

Is AI content bad for SEO?

Google says the way content is made is not the test. Its February 2023 guidance, Google Search's guidance about AI-generated content, written by Danny Sullivan and Chris Nelson, sits under the heading "Rewarding high-quality content, however it is produced." It also says, "Appropriate use of AI or automation is not against our guidelines." The same page draws the line at automation used to generate content primarily to manipulate search rankings, which it says is against Google's spam policies.

Google's spam policies on scaled content abuse describe what that looks like: "many pages are generated for the primary purpose of manipulating search rankings and not helping users." The first example on the list is "Using generative AI tools or other similar tools to generate many pages without adding value for users."

So the risk is unreviewed, mass-produced content that adds nothing a reader can use, and that risk applies to content written by humans too. Reviewed, sourced and useful content is the point of the exercise, and it is what our chain is built to produce. We looked closely at one version of the other approach in what "SEO on autopilot" actually buys you.

Google asks who, how and why

Google's page on creating helpful, reliable, people-first content asks site owners to think about who made their content, how and why. On the "how," it asks: "Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?" This article is our answer. You know which agents worked on it and which person approved it.

If you are comparing an AI SEO agency with one that writes everything by hand, the useful question is what review stands between the draft and your website. Ask to see the evidence of what was done, too. We wrote about what evidence shows that SEO work was completed, and it applies to any agency. Our own SEO service runs on the chain described here.

What to ask any agency about AI

Agencies that avoid AI have their reasons, and some do careful work by hand. Agencies that use it range from the careful to the careless. These questions sort them out, and they work whether or not you ever talk to us.

1. Which tasks in my project use AI, and which don't?
2. Who reviews AI-produced work, and is that reviewer someone other than whoever produced it?
3. Who approves the final version before it goes on my website or out under my name?
4. How are factual claims sourced, and can I see the sources?
5. Can anything be published, sent to my clients or charged to my ad budget without a person approving it?
6. Has the quality of your work changed since you started using AI, and what did you do with the time it saved?
7. Who will I actually talk to, and can I reach them directly?
8. What happens when the AI gets something wrong, and who catches it?

Listen for names and specifics. "Our proprietary AI handles it" is not an answer to question three. A good agency can tell you which person approves your work and how you reach them.

Talk to the people who sign off

Our agents do the research and the first drafts, and they check each other's work. Frank and Jana decide what reaches you, and they are the people you talk to. You can read more about both of them on the team page and see what we do on the marketing page. When you want to ask us the eight questions above, book a consult.

This article was prompted by a short video from TJ Robertson of TJ Digital, who makes the same case for his own agency. Worth a watch.

Common questions

What is an AI marketing agency?

An AI marketing agency uses AI tools for marketing work such as research, writing, design and checking. The useful difference between agencies is who reviews and approves that work. At Exhibit A Media, AI agents do most of the research, drafting and checking, and two people review the work: Jana Smith checks it against what the client needs, and Frank Alfano approves what ships. No agent publishes, spends money or emails a client on its own.

Does Exhibit A Media use AI to write its content?

Yes. Exhibit A Media's AI agents draft articles from a written brief, and separate agents review each draft for readability, sources and SEO before a person sees it. The writing agent never reviews its own work. Frank Alfano reads and approves each piece before it is published, and nothing goes out under a client's name without that approval.

Who reviews the work before it goes live?

Two layers review the work before it goes live. First, AI review agents check it: one for readability, structure and sources, another for keywords, metadata and links. They name fixes and leave the rewriting to the writer. Then a person decides. At Exhibit A Media, Jana Smith reviews work against what the client needs, and Frank Alfano approves what ships.

Is AI-generated content bad for SEO?

No, not by itself. Google's February 2023 guidance on AI-generated content says "Appropriate use of AI or automation is not against our guidelines," and it focuses on the quality of content rather than how it was produced. Google's scaled content abuse policy targets many pages generated mainly to manipulate rankings without helping users. Google says its systems aim to reward original, high-quality content, however it is produced.

Will AI replace marketing agencies?

Not the whole job. AI is replacing the generic work, the kind anyone can now get from a chatbot in the time it takes to type a request. What it doesn't do on its own is know your business, choose an angle that suits your clients, check its own claims or take responsibility for what goes out under your name. Agencies that keep people doing that work still have a job.

Should my marketing agency be using AI?

Your marketing agency can use AI well if it puts the time saved back into your work and keeps a person approving what ships. Ask which tasks use AI, who reviews the output, who approves the final version and how claims are sourced. An agency that answers with names and specifics is using AI carefully. One that can't name its reviewer is a risk.

Will I talk to a real person at Exhibit A Media?

Yes. At Exhibit A Media you talk to Frank Alfano or Jana Smith, the two partners who review the work. Frank approves what ships. Their phone number and email addresses are on their team pages at exhibitamedia.ca, and you can book a consult with Frank online. An AI agent may draft a reply to your email, and nothing is sent until Frank approves it.