When I was 15 I was building websites for local small businesses, mostly out of boredom and mostly for cash. I hosted them through a small hosting company, and the CEO there caught wind of what I was doing and hired me on the spot. That's how my whole programming career started - not a resume, not an interview, just a guy noticing I could already do the thing he needed done.
My first real job there was building a web-based billing system, and debugging clients' Perl and PHP and Apache setups while the hosting platform itself ran in Java, which meant I was definitely in over my head with OOP at 15. I went and found every CS resource I could - Stanford had just started posting professor syllabi online, which felt like finding a cheat code - and bought Stewart's Calculus and The Art of Computer Programming and a stack of Andre LaMothe game programming books I only understood half of. Then the dot-com bubble popped, the company lost its investors and most of its customers, and I lost that job. First lesson in how fast the internet can hand you something and take it back.
I bring that up because Perplexity has sent me traffic.
Not "someone Googled a thing and clicked a blue link" traffic. More like "an assistant summarized three vendors, picked two citations, and the buyer landed on a page I wrote two years ago" traffic. Same internet that hired a 15-year-old off the strength of a few HTML pages is now handing pages to buyers without a human ever typing a query into a search box. Different front door, same game underneath: something is reading your page and deciding what you're worth based on what's actually on it.
The old "rank and pray" playbook doesn't help much if the buyer never sees a SERP. They see a paragraph and a couple links, and they click one.
The new "SERP" is a paragraph
A lot of buyers are arriving through:
- A summary
- A comparison table
- A "top options" list with two citations
- A follow-up like "ok, which one works with X?"
The assistant is speed-running buyer research. It reads your category page, your competitor's FAQ, and three Reddit threads, then writes a neat little paragraph like it did the work.
If your content can't survive being compressed into a paragraph, the assistant will guess. And it'll guess wrong in the most annoying ways possible.
Not maliciously. Just mechanically.
If you lie, the assistant will quote the lie
The SEO problem and the fake-AI problem are related: both create a gap between what the page says and what the product does.
Now picture an assistant trying to summarize a company that calls spreadsheet cleanup "AI automation." It's going to call it AI automation because that's what the pages say. Buyers show up expecting magic. Support gets to explain why it's actually a bunch of hard-coded rules and a human checking edge cases.
If you're the honest competitor, your content has to say exactly what you do so you don't get lumped in with the spreadsheet theater.
Positioning still matters. Now it has to be quotable
The Crossing the Chasm framework is useful here. Not the MBA version. The "what do I actually put on the page" version:
Market category
State what category you're in, in plain words.
Bad: "We're an AI platform for operational excellence."
Better: "We're a document extraction API for insurance claims."
Competitive alternative
Name what people do without you.
- "Manual review in Zendesk"
- "Excel templates emailed around"
- "A BPO team retyping PDFs"
If you don't name the alternative, buyers will. Assistants will too. And they'll pick the laziest version.
Product alternative
Say what you replace.
If you're a smaller, specialized model, say that. If you're a rules engine with a model bolted on, say that too.
If you're vague now, you pay later. Every promise on the page becomes support work, sales cleanup, or an implementation surprise.
Write "answer-shaped" pages
I got handed a project once where the job was "build a better license plate reader." The readers on the market at the time needed to sit several meters back from the car to get a clean shot, which made them useless in a cramped downtown parking garage - there just wasn't room. Everyone's instinct was to chase better optics: bigger sensor, better lens, more resolution, more cost. Instead I used off-the-shelf cameras at lower resolution and mapped the features that actually mattered - make, model, shape, color, dents, stickers, whatever letters were visible on the plate. Got good results across every test garage we tried it in, without a single expensive camera.
Answer-shaped content works the same way. You don't need the fanciest page, the longest page, or the most "authoritative"-sounding paragraph. You need to map the right features - the ones a buyer or an assistant is actually trying to extract - instead of spending your budget on production value nobody's going to read closely.
Assistants like content that looks like an answer. Humans like it too, because nobody wants to read your brand manifesto just to learn if you support SSO.
Create pages like:
- FAQ pages that answer real buyer questions (pricing, data retention, deployment, evaluation)
- "X vs Y" comparisons with a table
- Implementation guides with real steps, failure modes, and how to fix it
- Definitions that don't read like a glossary written by an intern
This structure works because you can't bullshit your way through it:
What it is:
Who it's for:
What it replaces:
What it does not do:
What you need to make it work:
How we measure success:
Common failure cases:
If you can't fill in "what it does not do," you're not done.
Use AI for drafts. Use software for truth
I use AI to:
- Generate outlines for FAQ pages
- Turn messy notes into a first draft
- Refresh old posts so they match current positioning
I use simple tools for:
- Search Console and analytics (what actually gets found)
- Server logs (what actually gets crawled)
- CMS templates, schema, canonicals, redirects, sitemaps (deterministic stuff)
- Content inventory (yes, a spreadsheet, sorry)
I don't trust a model to tell me what Googlebot hit last week. I trust logs. Models can summarize the crawl story after the fact, but Search Console, server logs, and CMS data are the ground truth.
Keep the human parts human
This is why editorial standards matter. If your content is lying, the assistant will repeat the lie. Buyers will show up expecting the thing you don't actually do. Your support team gets to explain why the product doesn't match the promise.
Write so a human can decide and an AI can quote you without turning you into generic SaaS slop. That's the floor now.
If you want help, I run an AI-search audit: content inventory, positioning fixes, answer-shaped pages, and a crawl report that shows what actually gets indexed.
-Sethers
