Why I Verify Every Fact Before It Goes on a Site

Why do I verify every fact before it goes on a site?

Because one fabricated fact can undo years of trust. When my job is getting a site cited by AI answer engines, every claim on that page has to be true, or the citation becomes a liability. So I check each fact against a real source before it ships. It is slower, and it is the whole job.

I have published more than 350 articles on AI answer engines, schema, and E-E-A-T over the years, and the discipline that matters most is not clever writing. It is refusing to publish anything I cannot stand behind. That habit is unglamorous, but it is the reason clients trust the work.

Let me explain what verification actually looks like in my practice, and why I treat it as non-negotiable.

What does verify a fact actually mean in my workflow?

It means tracing every claim back to a real, primary source before it goes live. A statistic needs a named publication and a year. A product or event needs the vendor's own announcement. A quote needs the person who said it. If I cannot point to where a fact came from, it does not belong on the page.

In practice, I keep a simple rule: no claim without a source I have actually seen this session. Not a source I remember, not one that sounds right, but one I have opened and read. Memory is unreliable, and confident-sounding numbers are the easiest thing in the world to get subtly wrong.

This applies to my own numbers too. When I cite a result from my work, I use figures I can defend, like the ones behind my automation projects, not rounded-up guesses. The same standard I hold for outside sources, I hold for myself, which I wrote about in my post on designing a results section with honest numbers.

Why does this matter more now than it used to?

Because AI answer engines repeat what they find, and a wrong fact spreads faster than ever. When ChatGPT, Perplexity, or Google's AI Overviews pull a claim from your page, they can restate it to thousands of people. If that claim is false, you have not just misled one reader, you have seeded a lie into systems that echo it.

Search used to be more forgiving. A questionable stat sat on a page and mostly stayed there. Now a model can lift it, strip the context, and present it as settled fact in an answer. The blast radius of a single error has grown, which raises the cost of getting anything wrong.

There is a credibility angle too. AI systems and the people who read them increasingly value sources that are careful. Being the reliable one is a competitive advantage in a web flooded with confident, unchecked content. Accuracy is no longer just ethics. It is positioning.

What happens when a fabricated fact gets published?

The damage is disproportionate to the mistake. One invented statistic or made-up product detail, once caught, makes a reader doubt everything else on the page. Trust is not lost in proportion to the error. A single provable falsehood can poison an entire article, and sometimes an entire brand.

It can also invite a real response. If you attribute a fake figure to a named company, their communications team can notice, and a takedown request or a public correction is not a good day for anyone. The downside is not theoretical. Inventing a fact about a real organization is picking a fight you will lose.

I have seen unchecked stats travel from one blog to another until nobody remembers they were never true. That is the quiet danger. Fabrication does not always get caught immediately, but when it does, it discredits everyone who repeated it. I would rather publish less and be right than publish more and be the source of a lie.

How do I check a statistic before I use it?

I confirm three things: the source is named, the source actually published that figure, and I have seen it this session. A stat needs a real publication or organization behind it. If I cannot find the original, I do not soften it to "studies show." I delete it. A vague stat is just a fabrication in disguise.

The order matters. I do not write a number and then hunt for a source to justify it, because that tempts me to accept a weak match. I find the source first, then write the claim to fit what it actually says. Reverse-engineering a citation to support a number I already wrote is how honest people drift into dishonest pages.

When I use a figure, I name the source in the sentence, like "according to a 2026 SparkToro study" or a named report from Semrush or Gartner. That inline attribution does two things. It lets a reader check me, and it forces me to have a real source in the first place. If I cannot name who said it, that is my signal that the claim is not ready to publish.

How do I handle facts an AI tool gives me?

I treat every fact from an AI tool as unverified until I confirm it myself. Models like Claude, Gemini, and ChatGPT are useful for drafting and research, but they can state something false with total confidence. A number or a source that a model hands me is a lead to check, never a fact to publish.

This is the mistake I see most often now. People ask a model for a statistic, it produces a clean-sounding figure with a plausible source, and they paste it straight into a page. The problem is that the model may have invented both. A confident citation from an AI is not evidence, it is a hypothesis.

So I use AI to work faster, not to lower my standards. It can find candidate sources, summarize a page, or draft a section. But the verification step is mine, and it means opening the real source, whether that is Google Search Central for a search claim or a vendor's own newsroom for a product one. It does not get skipped because a machine sounded sure. I dug into this habit further in my post on how to fact-check AI-written content on a client site.

What do I do when I cannot verify something?

I cut it or clearly label it as unconfirmed. Those are the only two honest options. If a claim would make the piece stronger but I cannot find a real source, the piece runs without it. Holding back an unverified fact is never a loss, because the alternative is publishing something I might have to retract.

Sometimes a claim is genuinely uncertain, like a rumored feature that has not been confirmed. In that case I say so plainly, using words like rumored or unconfirmed inside the sentence. What I never do is assign a fake date or a made-up detail to make an unconfirmed thing read as settled. Dressing up a guess as a fact is the exact failure I am guarding against.

This restraint is hard in the moment. A specific number always feels more persuasive than an honest "I could not confirm this." But manufactured specificity is worse than honest generality, because it fakes authority you did not earn. When a client's reputation is on the line, I choose the version I can defend.

Does verifying slow me down too much?

Yes, it slows me down, and it is worth every minute. Checking sources takes real time, and there are days I wish I could just trust my memory. But the cost of verification is small and predictable, while the cost of a published falsehood is large and unpredictable. That trade is easy once you have seen the downside.

The speed cost is also lower than people expect. Most facts confirm quickly when you know where to look, and the ones that do not confirm are exactly the ones you should be suspicious of anyway. The friction is a feature. It stops weak claims before they reach the page, which is precisely its job.

Over time, verification even makes me faster, because it builds a library of sources I trust and claims I have already checked. The discipline compounds. What feels like a tax early on becomes a system that lets me write with confidence, knowing that everything on the page can survive scrutiny.

Why is this especially important for AEO and GEO work?

Because getting cited by AI is the goal, and a citation only helps if the fact behind it is true. My whole practice is built on making a site the trusted source an answer engine quotes. A fabricated stat does not just fail to help, it actively destroys the credibility I am hired to build. Verification is the foundation.

When a business hires me to improve its AI visibility, it is trusting me with its name in front of new audiences. If I put an unchecked claim on their site and a model repeats it, the failure lands on them, not on me. That responsibility is exactly why I hold the line on facts. Their reputation is the product.

There is a strategic point too. In a web where AI content is often sloppy and unverified, being rigorously accurate is how you stand out to both models and people. Correctness is not a constraint on the work, it is the work. I explored a related failure mode in my post on what to do when ChatGPT states wrong facts about your business.

What should you do next?

Adopt one rule before you publish anything: no claim without a source you have actually checked this session. Trace your stats, confirm your product and event claims against the vendor, and cut anything you cannot verify. It will slow you down at first and make your content far more trustworthy, which is the trade that pays off every time.

If you care about being cited by AI and read by people who trust you, this discipline is not optional, it is the whole game. If you want help building a site and a content process that answer engines can rely on, or a review of claims already on your pages, reach out. I am happy to walk through it with you. Let's connect.