Methodology

Layman's Terms reads a company's website and the coverage around it, then explains the company in plain English: what it does, what makes it different, and how to say it.

Why I built this

I've always had a passion for marketing and brand strategy. Good marketing connects people to products they don't know they need yet. It's matchmaking via translation (and distribution).

The Bay Area produces more interesting technology than anyone can keep up with. But most tech companies, especially B2B SaaS ones, have hard-to-understand and occasionally dull copy. As a PM & growth leader, I've often tried to evaluate a software product by visiting their website, only to leave the site to search for answers to specific questions (“Does Company X offer such and such functionality?” “How does it compare to such and such alternative?”) These companies make some of the most innovative products in the world, but their magic is often buried. You can't figure out what they do, what makes them different, and whether the product is worth your time, without scheduling a 45-minute demo.

I built Layman's Terms to address that for myself and for anyone else trying to keep up.

How it was built

I started with a list of questions a non-expert might actually want to know about a company, whether they're a prospect, a job seeker, an analyst, or a competitive researcher. That framing became the structure of the prompt.

The voice & tone came from my own opinions on what makes content easy and fun to read, then testing the prompt against real companies in deliberately varied situations: consumer vs. B2B, private vs. public, one product vs. multi-product, narrow audience vs. broad.

Each company exposed a weakness in the prompt, and the fix became the next version.

Adjustments I made along the way

When I started building v1 via Claude Code, the initial analysis took 28 minutes. I knew nobody would wait that long. The prompt now tells the model to read the website first, identify the gaps, then intelligently source what's missing (dynamic filtering). Tuning the number of searches, result depth, and how much text gets read per source brought the time down without hurting quality.

I also compared faster vs. higher-quality model versions, and surprisingly, the fast model produced better output. Same format, similar length and content, but punchier writing without losing substance.

Guidance I built into the prompt

• Every claim has to be traced to a reliable source: the company's own site, primary news, named analysts, court filings, etc. AI-generated comparison posts, anonymous blogs, and single social posts are never used. If a claim can't be backed, it doesn't appear.
• Don't shy away from depth, but explain in plain English. Technical terms and industry references are often important to include, but warrant a short explanation when first mentioned.
• Generate several options, then choose the best. For taglines, the prompt generates seven candidates internally and shows only the three that survive a strict ownability and clarity test. The reader should see the finalists, not all the drafts.
• When in doubt, use the buyer's words, not the analyst's. Copy should be written in the language the actual customer would use.
• Be generous through specific analysis, not vague flattery. The goal is to name what's genuinely hard, rare, or impressive, not to shower false praise. Critique, where it appears, should be direct and never snarky.

What makes this hard

• Copy is more art than science, and probabilistic models are better at producing averages than following nuanced rules. Take taglines. An ideal headline is both clever and specific. But a clever headline that doesn't tell you anything is worse than a boring-but-descriptive one. That line is subtle, and the model can sometimes miss it.
• It's hard for the model to treat headline, subheadline, and supporting copy as a system; getting one sentence right doesn't mean the whole page reads right.
• Models don't follow rules consistently, even hard rules baked into the prompt. Mistakes still creep in.
• Lastly, there's a structural issue: homepage copy changes constantly. Good growth teams are always testing new designs and messaging. Features, positioning, funding, and customer lists evolve. Every teardown is a snapshot in time.

What's still in progress

• This doesn't yet include user reviews, such as from Reddit, G2, or other sources. I may consider that for a future version.
• The tool reads a website and the public coverage around it. If you've used a product, you may know important things the homepage doesn't yet mention. In a future version, I may allow the user to include their own notes as input.
• The “why this matters now” section sometimes drifts into investor framing instead of buyer framing. This is partially addressed by a rule but not yet perfect.
• This doesn't account for brand personality / guidelines, which are foundational for good copywriting.
• Occasionally, some AI-style writing slips in. I'm continually tuning the prompt to remove this.

Have feedback or ideas for a future version? Drop me a note on LinkedIn!