This site publishes SEO and AI search news every day, and AI helps make that possible. That is exactly the kind of site Google's spam policy describes, if it's done carelessly: "using generative AI tools or other similar tools to generate many pages without adding value for users."

So we wrote down the process before we published a single post. Here it is, including the mistakes we caught while building it.

Step 1: A watcher reads the news, not the articles

Every morning a script reads the public feeds of nine sources: Google's Search Central blog, Google's search status page, Google's search product blog, Bing's webmaster blog, OpenAI's news feed (filtered to search topics), Search Engine Roundtable, Search Engine Journal, Semrush and Ahrefs.

It keeps only the headline, link, date and a two-line summary of each item. It never stores or rewrites anyone's article. It identifies itself honestly and only downloads a feed when it has changed.

One source is missing on purpose. Search Engine Land blocks automated readers, its feed included. We could find ways around that, but a block is a publisher's decision, so we respect it and read their site the normal way instead.

Step 2: It scores and groups the stories

The script gives each item points for how official the source is (Google's own blog beats a vendor blog), for topics that change what site owners do (core updates, AI Overviews, Search Console, crawling), and for freshness. It takes points away from webinars, recaps and evergreen "10 tips" posts. When several sources cover the same story, it groups them into one entry with several links.

The output is a ranked brief. On our first morning it held 28 stories.

A bug worth knowing about: our first version merged unrelated stories, such as TikTok ads and ChatGPT ads. The cause was one publication ending every headline with an author credit like "via @sejournal". Those shared words made different stories look alike. Stripping the credit fixed it. Small details in feeds matter.

Step 3: A person picks, using four questions

The score is a first sort, not the decision. On a busy news day we may publish up to three stories; on a quiet one, none. Each story has to pass four questions, in order:

  1. Does it change what a site owner should do or expect?
  2. Can we add something no one else has? Our own test, data from client sites, or a fix we found. If the answer is no, we skip the story, however high it scored.
  3. Is there a primary source? The original announcement, not another site's report. If a report is all there is, we say "reported" and link it.
  4. Have we covered it already?

On day one, the highest scoring story was about a search engine selling its index to AI companies. We passed: our only angle would have meant scraping that search engine's results, which its terms don't allow. A story the script ranked near the bottom, about a reported spike in Search Console clicks, became our first news post, because we could check it on twelve real sites. That story is here.

Step 4: AI drafts, a person checks every claim

AI helps turn our notes, data and fixes into a draft. It never supplies the substance. Every number in a post comes from a source we can show, and the editor checks each one before publishing.

We learned why this matters on our own sites. One of them published about 200 general blog posts over a few months, written with AI help. Of the 141 that showed up in Google over three months, 132 got zero clicks. Nothing was penalized; Google simply had better pages to show. More on that here.

Step 5: Automated checks before anything goes live

A post can't be published until it passes these checks, all run by script:

  • No client names. Clients are always described ("a Seattle apartment building"), never named. The script checks every post against our client list, including short names and domains.
  • House style. US spelling, and no em or en dashes.
  • Search basics. The title and description must fit what Google shows, or the post needs shorter versions.
  • Images. Every image needs alt text, and the build stops if one is missing.
  • Accessibility. Every page is tested against WCAG 2.1 AA at desktop and phone widths.
  • Scheduling. A post dated in the future stays hidden until its time, so nothing goes out early.

What we'd tell anyone using AI for content

  • Use AI for reading and drafting, not for knowing. It can sort 28 stories in seconds. It can't tell you which one your readers need or what you found on a client site.
  • Make "something only we have" a hard rule, not a goal. Skipping a story is a normal outcome.
  • Publish at a human pace. Our ceiling is three stories a day, and only when each one passes the four questions. Most days it is one or two.
  • Write your checks down as code. Rules people have to remember get forgotten on a busy day.
  • Say how you work. Our About page explains how we use AI, as Google suggests.