`> I'm the developer behind a privacy-first RAG chatbot for WordPress. This is a writeup from pagecoder.ai about a small tool I ended up building one weekend.
This one didn't start with a product idea. It started with an ad, on a Saturday morning.
Weekends are when I get to be a nerd for fun. No client work, just coffee, a croissant, and whatever rabbit hole the internet drops me into. This particular Saturday my feed was full of the same pitch, over and over: we'll get you into ChatGPT. Get cited by Perplexity. Show up in Google's AI answers. The future of search is here and you, my friend, are invisible in it.
I didn't buy anything. But I went looking for the price. And then the croissant went down the wrong pipe.
The business model was "hope you don't understand this"
Almost nobody sold it as "here's what's broken, here's the fix." They sold it as a relationship. Monthly retainer, content on a schedule, a dashboard that mostly reminds you to keep paying. The pitch leaned on one quiet assumption: this is AI, it's far too complicated for you, so hand it over forever and don't ask questions.
That's a business model built on the customer not understanding the thing. And the annoying part, as a developer looking at it, is that the expensive mysterious part is mostly just a measurement.
Most of "AI visibility" is measurable
I wrote down what it actually breaks into. It's not magic. It's about five questions, and most of them are things you'd check in a terminal:
-
Can AI even reach your site? The new crawlers (GPTBot, ClaudeBot, PerplexityBot) obey
robots.txtlike everyone else. Plenty of sites quietly block the exact bots they're paying someone to get them into. One-line check. - Can it read what it finds? Models read your page in chunks. If it's one undifferentiated wall of markup, it chunks badly, and badly-chunked content is hard to retrieve and cite.
- Does it know what you are? Structured data (JSON-LD), clear entities, consistent naming. This is how a model anchors "this page is about this entity that does this thing" instead of guessing.
- Can it actually quote you? You can literally test this. Hand a page to a model and ask what it understood and would cite. If the extraction is thin, so is your citability.
- Do you actually show up? Ask a search-capable model a real question in your niche and see whether you're in the answer, only in the sources, or nowhere.
None of that needs a subscription. It needs someone to run the checks.
So I built the audit, not the retainer
That's the tool. You give it a URL, it runs those checks, and it hands back a readiness score, a breakdown by category, and a list of specific findings, each with the evidence and the fix. Then you go fix them. That's the whole idea. If some fixes are worth outsourcing, outsource them to whoever you like. The diagnosis shouldn't cost you a salary.
The first URL I pointed it at was my own, because you don't ship a mirror without looking in it first. It found things. I fixed them.
One rule I built it around: no fake confidence
Most of these dashboards feel slippery because they show a guess as a fact. One confident number you can't interrogate.
So the scan has one rule: every signal says how sure it is. Measured, estimated, or not measured. If I actually fetched your robots.txt and read the rule, that's measured. If I'm inferring authority from public signals, that's estimated and labelled. If something genuinely can't be known from the outside, it says so instead of inventing a number to fill the box. Less impressive, more honest. That's the trade I want.
It's a loop, and it's honest about timing
A scan is only useful if you can tell whether your changes worked, so it's built as a loop, not a certificate. Scan for a baseline, make your fixes, scan again, diff.
For on-page readiness that loop is fast. Fix your robots.txt or your structure, re-scan minutes later, watch the score move. Visibility is slower and it says so. When you change your site, the AI answers don't update the same afternoon, because the models and search engines have to recrawl and reindex you first. Give it a couple of weeks before you expect the visibility probe to move.
The part no scan can fix for you
Here's the bit the ads skip. Most of your on-page readiness you can fix in a weekend. But whether a model actually trusts and quotes you is mostly not on your page at all. It's your reputation across the rest of the web: who links to you, who mentions you, whether you're a known entity or a stranger to the model.
No scan conjures that overnight, and neither does a monthly retainer, whatever the pitch says. What a scan can do is separate the easy on-page problems you fix today from the slow reputation ones that only real work over time moves. Knowing which is which, before you pay anyone, is half the battle.
Take it as a nudge, not gospel
The reason I built the measurement instead of reading opinions is that this stuff gets asserted a lot and checked rarely, and it's very checkable. If you're wondering whether an LLM can see your site, don't take a vendor's word for it, and don't take mine. Point something at a URL and read what the model actually makes of the page.
I build privacy-first tools for the web. This one is AI Visibility Scan - point it at a URL and see where you stand, free to try. My main product is RAG Chat, a privacy-first AI chatbot + search for WordPress. Need something custom built? Tell me what you need. No tracking pixels were used in this post.`
Top comments (1)
I appreciated how you broke down the concept of "AI visibility" into five measurable questions, particularly the point about models reading pages in chunks and the importance of structured data like JSON-LD for clear entity recognition. The idea of a readiness score with a breakdown by category and specific findings is really useful, and I like that you prioritized transparency by indicating the confidence level of each signal. Have you considered integrating this tool with existing SEO auditing tools to provide a more comprehensive analysis of a website's visibility and search engine optimization?