Garbage In, Garbage Out: What You Feed AI Decides What You Get Back
Picture two new employees starting on the same day. You hand the first one a sticky note: “Write something nice about our spring special.” You sit the second one down and explain who your customers are, what you charge, what you're proud of, and what you'd never say. Same skills. Same tools. Which one hands you work you can actually use? (That's an illustration, not a study.)
AI works the same way. It is a capable tool, and like a capable new hire, it can only work with what you give it.
What you feed AI decides much of what you get back, and the most useful things to feed it are the ones only you have.
In plain terms, “garbage in, garbage out” means vague, generic inputs produce vague, generic results, while specific details only you know give AI something real to work with.
Ipsos, a market research firm, tested 20 ads from 10 well-known brands in 2026, each made once by humans and once by AI. Its conclusion about the results: “There's no substitute for a strong brief.” It adds: “Creative outputs are only as good as the strength of their inputs” (Ipsos, 2026). A brief is simply the set of instructions you give before the work starts.
Similar Asks Get Similar Answers
Graphite, a growth agency, gave nine AI models the same prompts in a September 2026 analysis and compared the writing. Every model's writing looked more like some other model's than like a human's. Graphite notes that some of that similarity may come from the task itself rather than the models (Graphite, 2026).
A peer-reviewed story-writing experiment points the same direction. Writers who got AI-generated story ideas produced stories rated as more creative, but those stories were more similar to each other (Doshi and Hauser, 2024).
Here's my read, and it's an inference, not something either study tested for business owners. When many people send AI the same kind of ask, they get back the same kind of answer. “Write a friendly post about our spring special” hands the tool nothing a competitor couldn't hand it too.
What Only You Can Feed It
AI can't know four things unless you tell it. The specifics of your business: your services, your neighborhoods, a real price range, how long you've been doing this. Who you're talking to. How you sound. And what you would never say.
That list is the difference between a brief and a wish.
Should You Put Everything in the Prompt?
Feeding AI well doesn't mean feeding it everything. In a study published in Transactions of the Association for Computational Linguistics, researchers tested language models on questions that required finding relevant information inside long inputs. Performance was often highest when the relevant information came at the beginning or end of the input, and it dropped considerably when the information sat in the middle (Liu et al., 2024). For GPT-3.5-Turbo, accuracy on multi-document questions could fall by more than 20%, and in the worst case it landed below the model's accuracy with no documents at all (Liu et al., 2024).
Two limits. The study tested the models available at the time, on long-document questions rather than a short business brief, so I'm not claiming the same drop in your prompt. My read: keep your brief short and put what matters most first.
Can You Teach AI to Write Like You?
Showing AI examples of your own writing is a common fix. A study in Findings of EMNLP 2025 tested how well models could imitate a person's writing style from a few of that person's own samples, using more than 400 real authors across news, email, forums, and blogs. The models approximated style reasonably in structured formats like news and email, but struggled with nuanced, informal writing in blogs and forums (Wang et al., 2025).
For a local business, here's my read: samples are worth giving, especially for emails and service descriptions. For casual social posts, expect to edit the result until it sounds like you. I'll cover that review habit in the last piece in this series.
A Five-Minute Brief
Before you ask AI for anything, write these five lines. They're my own recommendations, built from the research above, not a tested formula.
Say who it’s for. One sentence about the reader: who they are and what they’re worried about when they find you.
Give it facts only you know. Your services, your service area, a real price range, a recent job (with the customer’s permission), how long you’ve been at it.
Show it how you sound. Paste two short things you wrote yourself, like an email or a text to a customer, and ask it to match that voice.
Tell it what to leave out. Stock closing lines, claims you can’t back up, anything you would never say out loud.
Save it. Write the brief once and reuse it. Update it when something changes, like a new service or a new price.
How We Set Up AI for Clients
Every client starts with an onboarding session. Before we build anything, we learn about you and your business: what you offer, what you charge, where you work, who your customers are, how you talk to them, and what you would never say. That information shapes whatever we set up for you, whether it's a chatbot, a voice agent, your social posts, your website, or your follow-up messages. AI can only be as accurate as what it has been given, so the inputs come first and the output second. That isn't a guarantee of results. It's the order I insist on.
One Thing to Check Right Now
Pull up the last prompt you gave AI. Count the details in it that only your business would know. If the count is zero, add three before you run it again: one fact, one sample of your voice, and one thing to leave out.
Frequently Asked Questions
What does “garbage in, garbage out” mean for AI?
It means the quality of what you get depends on the quality of what you give. A vague ask with no real details tends to produce generic output. Ipsos's 2026 ad test put it this way: “Creative outputs are only as good as the strength of their inputs” (Ipsos, 2026).
Should I put everything in the prompt?
Not necessarily. Research on long inputs found that models did best when the relevant information came at the beginning or end and worse when it was buried in the middle (Liu et al., 2024). That study tested earlier models on long-document questions, so treat it as a reason to be concise and lead with what matters most.
Can I teach AI to write like me?
Partly. In a study of more than 400 real writers, models given a few samples approximated style reasonably in structured formats like news and email, but struggled with informal blog and forum writing (Wang et al., 2025). Give it samples, then read the result out loud and fix whatever doesn't sound like you.
What if I don't have time to write all this every time?
Write the five lines once, save them, and reuse them. Update them when something changes in your business, like a new service or a new price.
Here's the Plan
You don't need a better tool. You need to hand the tool what only you know.
I'm Paul, founder of Get Local Visibility. I help local business owners get found, get chosen, and get every lead followed up with.
If you'd rather have someone set up your AI with your real business details and your voice from day one: the research keeps pointing to the brief, not just the tool. Book a free, zero-pressure strategy session with me today. If you want us to build it for you, great. If not, you walk away with a free roadmap.
This piece is part of our AI series, “AI Isn’t the Problem. How You Use It Is.” Start with the overview: The Marble and the Chisel: Why Bad AI Content Usually Isn't AI's Fault. Also in the series: The Em Dash Problem: What Actually Gives Lazy AI Writing Away and Why So Many AI Ads Look Fake, and What Actually Fixes It.
