Best Practices

2 Hacks to Get More Reliable Answers from AI

Two ways to protect your information diet from AI

Summary

AI answer engines are mostly pulling from 15 sources. This guide covers why GEO is driving that concentration, and two ways to keep your own research robust

Overview

Before you start

Why AI is limiting our information diets

The web used to run on SEO, content built to rank in a search engine that returned dozens of links, and you picked which ones to trust.

AI engines (Claude, ChatGPT, any model), on the other hand, scrape information from the web differently.

This is GEO, generative engine optimization — structuring online content so that AI chatbots can easily find, understand, and cite brands in their direct answers.

The problem with this is a handful of sources either have deals with OpenAI to appear in search results, or have mastered structuring data for AI search results. There are 15 websites, actually, that account for 68% of AI results.

You can protect your information diet in two ways detailed below.

How does search work in AI?

When you ask AI a question, it answers, but the sourcing happens somewhere you can't see.

Here's what's actually going on underneath that.

A June 2025 Semrush study of 150,000 AI citations across 5,000 keywords found that 40% of them pointed to a single source: Reddit.

This is how the internet is changing: The web used to run on SEO, but now, AI engines don't return links to sort through. They return one synthesized paragraph, built from whichever sources the model already leans on.

This increasingly means the same handful of platforms get cited, every time.

Part of why Reddit specifically is so dominant is because it's a licensed source. That's the theory why Reddit shows up disproportionately in answers.

None of this is a reason to distrust AI outright.

But you should stop treating its answers as neutral by default; the same way you wouldn't take one Reddit thread as 100% accurate.

If a narrow set of sources is driving what you see, it's worth building guardrails around your information diet.


Step 1: Exclude Low-Quality Sources in Your Prompts

When you're chatting with an AI tool directly, add an instruction to your prompt telling it what to leave out:

Exclude Reddit from your results, and only pull from real articles, reputable outlets, or academic journals
Exclude Reddit from your results, and only pull from real articles, reputable outlets, or academic journals

This forces the model to reach past its default, most-convenient sources and actually draw from reporting and primary data instead.

Caution: This only works for the exact conversation you're having.

It doesn't change what the model defaults to next time.

Step 2: Use a Research Skill that Enforces Source Rules Automatically

A skill is a rule that can't be skipped because it's encoded in your model.

I built a /research skill that does this automatically.

Before it searches anything, it checks a maintained exclusion list. Every source it does use has to clear one of three tiers — reputable news, academic or institutional, or named independent expertise — or it gets discarded.

The output isn't a paragraph you have to fact-check after the fact.

How it works: Download the skill .md file directly

Caution: A research skill is only as current as its exclusion list.

How does search work in AI?

When you ask AI a question, it answers, but the sourcing happens somewhere you can't see.

Here's what's actually going on underneath that.

A June 2025 Semrush study of 150,000 AI citations across 5,000 keywords found that 40% of them pointed to a single source: Reddit.

This is how the internet is changing: The web used to run on SEO, but now, AI engines don't return links to sort through. They return one synthesized paragraph, built from whichever sources the model already leans on.

This increasingly means the same handful of platforms get cited, every time.

Part of why Reddit specifically is so dominant is because it's a licensed source. That's the theory why Reddit shows up disproportionately in answers.

None of this is a reason to distrust AI outright.

But you should stop treating its answers as neutral by default; the same way you wouldn't take one Reddit thread as 100% accurate.

If a narrow set of sources is driving what you see, it's worth building guardrails around your information diet.


Step 1: Exclude Low-Quality Sources in Your Prompts

When you're chatting with an AI tool directly, add an instruction to your prompt telling it what to leave out:

Exclude Reddit from your results, and only pull from real articles, reputable outlets, or academic journals

This forces the model to reach past its default, most-convenient sources and actually draw from reporting and primary data instead.

Caution: This only works for the exact conversation you're having.

It doesn't change what the model defaults to next time.

Step 2: Use a Research Skill that Enforces Source Rules Automatically

A skill is a rule that can't be skipped because it's encoded in your model.

I built a /research skill that does this automatically.

Before it searches anything, it checks a maintained exclusion list. Every source it does use has to clear one of three tiers — reputable news, academic or institutional, or named independent expertise — or it gets discarded.

The output isn't a paragraph you have to fact-check after the fact.

How it works: Download the skill .md file directly

Caution: A research skill is only as current as its exclusion list.

Download Skill (.md)

CHEAT SHEET

Cheat Sheet


SEO

GEO

Goal

Rank in a list of links

Be the source an AI answer cites or synthesizes from

Success metric

The click

The citation

What it optimizes

A page's search ranking

Content and brand presence for inclusion in AI-generated answers

What brands have to do

Keyword targeting, backlinks, technical site health, page speed

Emerging, less settled tactics: structuring content to be easily extracted and cited, building topical authority a model would recognize

REMAINING QUESTIONS

Remaining Questions

Isn't excluding Reddit just throwing away real user insight?

No, the fix isn't "never use Reddit," it's "don't let it be the only thing an AI answer quietly leans on."

The research skill doesn't ban community sources outright; it just requires them to be named, credentialed, and tiered like everything else, instead of showing up by default because they're the easiest thing for a model to cite.

Won't a research skill slow me down when I just need a quick answer? Sometimes, yes.

That's what Step 1 is for. The exclusion prompt is the fast path.

How do I know the sources the skill does use are actually more reliable? Because the skill labels its own confidence instead of asking you to trust it blindly.

A claim relying on one source gets marked single-source, not presented as settled fact. The point isn't that every cited source is perfect.

But that you have a system that tells you when it isn't sure, which is more than most AI answers do right now.