Using an AI writer will not get your site flagged. Publishing pages that add nothing will. Run every draft through the ten-minute test below, publish only what passes, and check Search Console a week later to see if Google listed it.
AI is allowed. Mass-produced filler is the problem.
Google's guidance says appropriate use of AI or automation is not against its guidelines. What its spam policies target is scaled content abuse: producing many pages mainly to influence rankings instead of help people. AI makes that easy to do, but you can do it by hand too.
So the question is not "did a model write this?" It is "does this page give a reader something they cannot get from the ten pages already ranking?" If yes, the tool does not matter. If no, the tool is just how you made a junk page faster.
As a solo founder with a day job, you have a built-in advantage here. You cannot publish fifty pages a week, so you will not. Publish fewer pages and make each one carry something only you have.
Raw AI drafts fail in five predictable ways
The same flaws show up again and again in unedited drafts, and each one has a price:
| Flaw in the draft | What it costs you |
|---|---|
| Claims with no reliable source | Readers and AI answer engines stop trusting the page |
| Generic advice found on dozens of other pages | Nothing to rank on, no reason for anyone to cite you |
| Summaries with no new analysis | The reader learns nothing they could not get elsewhere |
| Sections added only to hit a keyword | Padding that thins out the page |
| Repetitive template phrases and vague sentences | The text reads as machine-written, as this breakdown of why AI content fails puts it |
Human writing can be thin or wrong too. The fix is not to hide that you used AI. The fix is review and evidence.
The ten-minute publish test: five checks, all must pass
This is the part to save. It is built from published guidance on editing AI content and adding experience to AI drafts. Set a ten-minute timer and run it before anything goes live.
- One problem. Can you state, in one sentence, the single reader problem this page solves? If not, do not publish.
- One thing only you have. Is there at least one example, screenshot, number, quote or mistake from your own work? If you have nothing for this topic, pick a different topic.
- Every stat traced. Find the original source for each number. If you cannot find it, delete the number.
- Names, dates and quotes checked. Confirm spellings and dates, confirm any quote is real and credited to the right person, and confirm any study you cite exists and says what your page says it does.
- Generic lines cut. If a sentence could sit on any site about the topic, rewrite it with a specific detail or delete it. Remove filler words like "unlock" and "delve".
Then read the draft out loud and cut 10 to 20 percent. One editing guide recommends this clarity pass. On a 1,200 word draft that means landing between about 960 and 1,080 words.
Pass all five and publish. Fail one and fix it. Fail two or more and the draft is not an article yet, it is notes.
Add your own material first, then let the AI fill the gaps
Most people do this backwards. They prompt for a full article, then try to sprinkle in some personality. Flip it.
Start with what you know: the setup you ran, the mistake that cost you a weekend, the decision rule you actually use. Write those down as rough bullets. Hand them to the AI as the spine of the draft and ask it to fill in the explanation around them. The experience signals that matter are the ones a model cannot make up: screenshots, data from your own project, quotes, and the mistakes you made.
If you cannot fill the spine, that is your answer. You do not have a page here yet. Go build something, break something, then write about it.
This also fits a day-job schedule. Capturing three bullets after a build session takes five minutes. Staring at a blank page on Sunday night takes an hour. If you are still working out where the hours come from, the side project schedule I use as a template covers the time-boxing.
Check indexing a week after you publish
A page can pass your test and still not get listed. Google may crawl a page and decide not to index it. Search Console calls that status "Crawled - currently not indexed", and it means the page will not appear in search results.
Here is how to look, free:
- Open Google Search Console.
- In the left menu, go to Indexing, then Pages.
- In the "Why pages aren't indexed" table, find the row "Crawled - currently not indexed".
- Click the row to see every URL with that status.
- Inspect one URL to see what Google saw.
If your page is on that list, do not just resubmit it. The listed causes include thin content, overlap with a stronger page on your own site, unclear canonical signals, weak internal links, an unstable server and rendering differences. The usual cure is a mix of internal linking, content upgrades and technical cleanup, not a technical tweak alone.
My order of attack, when X, do Y:
- Page is short or generic: add your own material first. Compare it to the pages that rank for your search and see how deep they go. If every top result runs past 3,000 words with real depth and yours has a few hundred, it likely reads as thin. Use that as a comparison for your search, not a word count to hit.
- No other page on your site links to it: add links from two or three older pages. A page with weak internal links can look unimportant.
- Content and links are fine: only then go hunting for canonical or server problems.
Example: one page a week on a 90 minute budget
Example: say you publish one page a week and you have 90 minutes for it. These numbers are illustrative, so change them to fit your week.
| Step | Minutes |
|---|---|
| Write three bullets of your own experience | 10 |
| Have the AI draft around them | 15 |
| Rewrite generic lines, add one real example | 30 |
| Run the ten-minute publish test | 10 |
| Read aloud and cut 10 to 20 percent | 10 |
| Add 2 to 3 internal links from older pages | 5 |
| Publish, then log the URL to check in a week | 10 |
That is 90 minutes and one page that earns its place. Compare that with sixty pages in a weekend that all say the same thing as everyone else. The first plan builds a site Google can trust. The second one looks exactly like what the spam policies describe.
When your week gets squeezed, skip a week. Do not skip the test.
Internal links and community posts do the work while a page waits
A new page on a small site does not get listed on day one. While it waits, send it real readers. Link it from your older pages, and if you are active in communities, share the useful part, not the link. The Reddit rules for solo founders cover how to do that without getting banned.
If you run agents to do some of this drafting, keep them boxed in. The portfolio write-up on how I run small apps with AI agents shows where an agent fits in a one-person workflow.
Do this in the next ten minutes
Pick the last AI-assisted draft you have not published, or the last page you did publish. Run the five checks on it right now. If it fails two or more, move it back to notes. If it is already live, open Search Console, go to Indexing, then Pages, and see whether it is on the "Crawled - currently not indexed" list.
Got a manual job you want automated?
If there is a repeatable task eating your hours, tell me what it is and I will tell you honestly whether a tool can handle it. Email Me.
