crontent

how to avoid spam email

Crontent

One of the weirder things in AI search right now: the same page can get cited by ChatGPT and vanish completely in Google AI Overviews. Eugen Ullrich tracked eight B2B projects for a month and found four out of five cited domains showed up in one engine only.

That sounds unrelated to spam email. It isn't. If you want to avoid spam email, the first job is to stop publishing the kind of thin, generic pages that look interchangeable with everything else flooding the web. The new AI citation data gives you a cleaner way to spot that problem.

How to avoid spam email when AI search rewards different pages in different engines

Write pages that answer one real question with clear facts, then test whether different AI engines actually cite them. Ullrich's data shows ChatGPT, Perplexity, and Google AI Overviews barely share source pools, so one dashboard won't tell you whether your page looks useful or just looks like filler.

Spam email usually follows the same bad pattern as spammy content. You blast the same message everywhere, with tiny wording changes, and hope volume wins. AI search is pushing the opposite way. The arXiv GEO paper looked at 602 prompts and 21,143 valid search-layer citations and found the engines don't just cite different amounts. They also use sources differently. Perplexity cites more sources per prompt. ChatGPT cites fewer, but the pages it does cite tend to have higher influence on the answer.

If you're trying to stay out of the spam bucket, that matters. Don't make twenty near-copy pages. Make a few pages that are worth citing because they have:

  • a clear claim
  • specific facts or numbers
  • obvious structure
  • one job per page
  • language a buyer would actually search

That's a better filter than asking whether one "AI visibility" tool says you're up this week.

One prompt set won't tell you if your content feels spammy or useful

A single prompt report only tells you how one wording performed on one day in one engine. The Claude Search work from NotPeople makes that painfully clear. In its Q1-Q2 2026 audits, Claude's overlap with Perplexity-cited sources sat below 40 per cent.

So if you test only one engine, or one set of prompts, you can talk yourself into a false win. Your page might look great in Perplexity and dead in Claude. Or it might get picked up by Google AI Overviews because it already fits Google's search index patterns, while ChatGPT ignores it.

That matters for avoiding spam email because spammy marketing usually shows up when you stop checking what people actually respond to. You start mass-producing copy instead of checking whether each message earns attention.

For a small SaaS team, the better weekly habit is simple:

  1. Pick 5 to 10 important prompts buyers use.
  2. Check them in ChatGPT, Perplexity, Google AI Overviews, and Claude.
  3. Log the cited URLs, brand mentions, and page types.
  4. Note whether docs, comparisons, stats pages, or blog posts win.
  5. Rewrite or kill pages that never get picked up anywhere.

That gives you evidence. Evidence keeps you from sending the content version of spam email.

Pages that avoid the spam label usually do one useful thing really well

High-influence pages tend to be longer, more modular, and more semantically aligned with the prompt, according to the arXiv measurement framework. That's a dry way of saying the page is easy to scan, sticks to the topic, and gives the model material it can actually use.

That is also how you avoid sounding like spam in email, on landing pages, and inside AI search. Pick one intent and satisfy it cleanly.

For B2B SaaS, the page formats worth testing first are usually:

  • comparison pages with concrete tradeoffs
  • docs that explain a task step by step
  • stats or benchmark pages with original numbers
  • use-case pages tied to a real workflow

Ullrich shows why format testing matters. If source overlap is low, a winner in one engine may not travel. NotPeople adds another layer: Claude treats the web more like live retrieval than a ranked index, so pages that depend on old SEO habits may not carry over.

If your page is vague, padded, or built from recycled AI copy, it will fail this test fast. It won't earn citations consistently across engines because there's nothing distinct to retrieve.

The practical fix is to publish less, log more, and stop sending copy nobody asked for

The cleanest way to avoid spam email is to stop acting like a spammer in your content. AI search now makes that visible. Different engines pull from different source pools, cite pages at different rates, and reward different formats, per Ullrich, the arXiv study, and NotPeople's Claude audit.

So don't chase one score. Track engine by engine. Keep a small prompt set. Watch cited URLs, referral traffic, crawlability, and page format. Then cut anything that reads like it could have been sent to everyone.

If a page could double as spam email, don't publish it. If it teaches one useful thing with real evidence, keep going.

Sources