Building a tool that audits websites + writes the cold emails for you (from someone who's been doing this manually for months)
I run a small web agency doing landing pages for local service businesses (cleaning, junk removal, that kind of niche). The actual work isn't the hard part — finding clients is. Every cold email I send starts with me manually checking a site's PageSpeed score, poking around for what's broken, then writing something personal enough that it doesn't read like spam.
That manual process is what I'm trying to build into a tool now.
The idea: paste in 5-10 business URLs → it crawls the site, runs a real technical/design audit (contrast, mobile layout breaks, tap targets, PageSpeed, the works), grabs a screenshot of the single worst problem as visual proof, tries to find a real decision-maker's name from the About page, and then writes 3 different cold email variants that actually reference the specific thing wrong with their site — not generic AI fluff.
No business search, no Google Maps scraping, no sketchy data sourcing — you bring the URLs, it does everything after that.
Where it actually is right now: rough. Got a real preview running last night for the first time after a couple months of on-and-off building, and honestly it's maybe 10% of where it needs to be. UI's a mess, performance is slow, the crawler isn't isolated per user yet so it chokes under any load, the screenshot proof-boxes are drawing in the wrong spots, contact finder is basically just guessing names off the page right now. Real list of problems, not humble-bragging.
But it's the first time I've had the full pipeline actually run end to end, and that felt good after chasing this idea for months. Now going through and fixing each piece properly — architecture, UI, the actual crawling logic — one at a time.
Posting here mostly to start being honest about the process instead of building in a cave. If you've dealt with the "finding clients is the actual bottleneck" problem, or built anything in this audit/outreach space, I'd love to hear what you ran into.
The "will AI take my job" market splits into meme quizzes that disclaim themselves out of seriousness, and paid reports that hide their methodology behind the same five recycled citations. I'm building the honest one.
AIRA is a free AI job-risk assessment with a fully public methodology — the results page shows you line by line how your score was composed. The business model is a paid personalised Kit (waitlist for now; checkout is next on the roadmap).
Solo founder in Adelaide, built with Next.js + Supabase + Claude doing most of the code under spec-driven review.
Live: useairanow
The bet I'd love feedback on: that transparency converts better than urgency in a category built on anxiety. If you run the score — did it feel right, and did the breakdown make you trust it more or less?
the "transparency over urgency" framing is smart — most tools in the AI-anxiety space lean hard into fear because it converts short-term, but it also means everyone's numb to it by now. a score you can actually inspect line-by-line is a different kind of hook entirely, since it's earning trust instead of borrowing urgency. I'm building something with a similar principle at the core (never present a guess as confirmed fact, always show your work) so this resonates — curious if you're seeing that trust translate into people actually joining the waitlist, or if transparency alone isn't enough to convert without more urgency mixed in?
I'm curious what convinced you the real bottleneck was producing personalized outreach rather than deciding which businesses are actually worth contacting.
Did your manual process consistently show that diagnosis was the expensive part, or that turning it into credible outreach was?
Honestly, it wasn't one or the other — it was that both steps were manual and slow, and skipping either one made the email worse.
Deciding if a business was even worth contacting was real work: pulling their PageSpeed score, checking if they had a booking/contact flow, seeing if the site looked maintained. If I skipped that and just fired off generic-personalized emails, reply rates were bad — people can tell when you didn't actually look at their site.
But the diagnosis alone wasn't the email either. Knowing "this site is slow and has no CTA" doesn't write itself into something a business owner reads and thinks "oh, they actually looked at my site." Turning the diagnosis into something that reads as credible and specific — not just "I noticed your website could be improved" — was its own separate time sink.
So the tool's trying to compress both: the "is this worth my time" check and the "make it sound like a human looked at this" step. If I only automated one of those I don't think it'd move the needle much — I've basically tried that manually (audit without good copy, and decent copy without a real audit) and both versions underperformed.
Appreciate the context.
Would be interesting to continue the conversation as you learn what part of the workflow creates the most value for users.
What's the best email to reach you on?
ahmadhasan.design@gmail.com — happy to keep chatting there too
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.