Isaac · Creative Director, Aphelion
Updated October 23, 2026 · 8 min read
AI is useful in marketing where a draft, a summary or a classification saves a human from starting at zero. It is unreliable where the output must be factually correct without someone checking it. Most of the value sits in the first category and most of the disasters in the second.
Where it genuinely helps
Turning a rough brief into a first draft. Summarizing a pile of customer feedback into themes. Classifying LeadSomeone who showed interest in what you sell and left their details or messaged you, but has not bought yet. or messages by intent. Generating variations of ad copy to test. In all of these a human still decides, and the AI removes the blank-page cost rather than the judgment.
Where it fails, confidently
Anything requiring a specific fact it was not given: your prices, your availability, a statistic, a competitor's claim. It will produce something plausible and well-written, which is precisely what makes the error expensive. Published content built this way is how businesses end up quoting numbers they cannot back.
The problem with fully generated content
Search engines are not the main issue: readers are. Text produced without anyone who knows the business reviewing it tends to be technically fine and completely interchangeable with a competitor's. It says nothing only you could say, which is the one thing content has to do to be worth publishing.
What our own stack looks like
We use Claude for drafting and for the conversational agents we build, automation tooling to move data between systems, and AI-assisted development to ship interfaces faster. What we do not do is publish anything a person on the team has not read and corrected, because the failure mode above is not theoretical.
What AI did not change
Strategy, positioning, and knowing which customer you are actually for. A tool that writes faster makes a clear strategy cheaper to execute and a confused one cheaper to broadcast. Teams that were unclear before mostly got louder, not better.
How to evaluate an agency that says it uses AI
Ask what a human still checks before anything is published, and ask to see something they built with it rather than a description. Using AI is not a differentiator anymore; having a process that catches its mistakes is. That question separates the two quickly.
Frequently asked questions
It can write useful drafts. Publishing them without someone who knows your business reviewing is where the problems start, because the text sounds fine even when the facts are wrong.
The public stance centers on the quality and usefulness of the content rather than how it was produced. In practice, generic content performs poorly regardless of origin, and unreviewed generated content tends to be generic.
It depends on the bottleneck. If the problem is producing content, a writing tool. If it's handling inquiries, a conversational agent. Buying several before knowing which one hurts is the most common way to spend without results.
It replaces tasks, not judgment. What did change is the expectation: if an agency bills for production hours that now take minutes, that model is under pressure. The value moved toward strategy and measurable execution.
Want this handled instead of researched?
Find out in a few minutes what to fix first in your business, or tell us what you're working with and we reply within 24 hours.
Marketing agency in Tijuana
