DEV Community: Veritas Links

LLMs are non-deterministic and your visibility dashboard is pretending they aren't

Veritas Links — Mon, 27 Jul 2026 18:40:05 +0000

A language model samples. Ask it the same question twice and you get two answers. Most monitoring tools ask once, record the draw, plot it on a dashboard, and call tomorrow's second draw a trend.

We tested how bad this gets. Fifteen identical runs of one prompt against one company. Four models stayed inside ranges of 44 to 79 points. Perplexity's sonar moved 232 points across the same fifteen runs and contradicted itself twice in a single afternoon. Nothing changed between runs except the sampling.

If your metric is one draw, a trend line built from it cannot separate a real movement from the model's own variance. We report the distribution instead of the point. Methodology and the reliability study are written up here https://veritaslinks.com/compare/profound-vs-veritaslinks

VeritasLinks launches GEO Analysis 3.0 — and it’s time to stop measuring the wrong thing.

Veritas Links — Mon, 22 Jun 2026 21:29:18 +0000

Almost every post and tool in the SEO/GEO space tells you the same story:
“Look at your citations. Look at your mentions. Look at your share of voice.”
That’s bullshit.
Citations are not recommendations.
A model can know your brand, list your services, mention you among competitors — and still never recommend you to potential customers.
We’ve seen it dozens of times: companies with near-perfect citation rates across 100 queries get completely ignored when the LLM is asked “Who do you recommend?” or “Which company should I choose?”
And yet, many GEO/AEO agencies and tools continue to sell the illusion of progress by optimizing for citations.
We took a different path.
With GEO Analysis 3.0, VeritasLinks focuses on what actually matters:

How LLMs truly perceive your brand
Whether they recommend you — or your competitors
Why they make that choice
What hidden objections, trust gaps, or perception issues are holding you back

We still track citations (it’s one of 50+ signals), but it’s no longer the main metric. The new version is built to understand the reasoning behind the model’s decisions.
Because in the age of AI, being known is not enough.
You need to be the one that gets chosen.
If you’re tired of vanity metrics and want to see the honest truth about how AI systems see your company, we made this analysis free to try.
→ Run your free GEO Analysis here: https://veritaslinks.com
Compare it with what other tools show you.
You’ll immediately feel the difference in depth and honesty.
Welcome to Version 3.0.

The GEO Industry Is Measuring The Wrong Thing

Veritas Links — Thu, 18 Jun 2026 23:56:39 +0000

Over the last year we've watched dozens of GEO tools appear almost overnight.

The 99.9% of them promise roughly the same thing.

Track citations.
Monitor mentions.
Measure visibility inside ChatGPT, Gemini, Claude, Perplexity and whatever model launched last Tuesday.

And honestly, we think the entire industry is becoming obsessed with the wrong metric.

Everyone is counting citations. But almost nobody is measuring recommendations. And those are not the same thing!

Imagine you're a founder and you ask your favorite LLM:

"Who are the major players in AI visibility analysis?"

The model responds:

Company A, Company B ..., Company ZYX"

Congratulations.

You got cited. Maybe.

Now imagine a potential customer asks:

What solution would you recommend for a growing SaaS company that wants to understand how AI perceives its brand?

Suddenly the answer changes.

Maybe Company A disappears.

Maybe Company C becomes the primary recommendation.

Maybe the model only recommends two vendors despite mentioning ten elsewhere.

That's the moment that actually matters.

Not whether your brand exists somewhere inside the model's response. Whether the model trusts your brand enough to recommend it.

The GEO industry has quietly started treating these two things as if they were interchangeable.

They're not!

A citation simply means a model knows your company exist.

A recommendation means the model has formed a positive enough opinion to put its reputation behind you.

That's a much higher bar.

And once you start looking at AI systems through that lens, some strange things begin to happen.

We've seen companies with excellent citation visibility that almost never get recommended.

We've seen companies with relatively few mentions become default recommendations for entire categories.

We've seen brands that appear in informational prompts disappear completely when buying intent enters the conversation.

That's because recommendation behavior is driven by something much deeper than raw visibility.

The model is evaluating confidence.

Can it explain what this company does? Does it understand the category? Does it see trust signals? Can it differentiate the company from competitors? Has the brand established enough authority across the information ecosystem for the model to feel comfortable recommending it?

Most GEO platforms never answer those questions.

They stop at citation tracking.

At VeritasLinks, we became interested in the layer underneath.

The layer where perception is formed.

That's why our reports don't simply count mentions across models. We analyze how models describe a company, what objections appear repeatedly, which competitors are positioned more favorably, what trust signals influence responses, and ultimately whether recommendation probability changes as prompts become more commercial and decision-oriented.

Because the future isn't going to be won by the companies that get mentioned the most.

It's going to be won by the companies AI systems trust enough to recommend.

If you're only measuring citations, you're measuring awareness.

If you're measuring recommendations, you're measuring influence.

Those are very different games.

If you're curious how AI models currently perceive your company, you can run a free analysis through VeritasLinks or learn more about how we approach AI perception and recommendation analysis here:

How to Check Brand Visibility in ChatGPT | Free AI Brand Analysis

Learn how to check whether ChatGPT mentions, recommends, or misunderstands your brand. Analyze AI visibility, trust signals, competitor positioning, and brand perception for free.

veritaslinks.com

For a deeper look at how our GEO analysis works, including source evaluation, trust signals and competitive positioning:

Best GEO Tools in 2026: Honest Comparison of the Top Generative Engine Optimization Platforms

Compare the best GEO tools in 2026 including VeritasLinks, Profound, Otterly, Brandwatch, Meltwater, Ahrefs and SEMrush. Learn which platforms actually improve AI visibility and AI recommendations.

veritaslinks.com

Because being mentioned is nice.

Being recommended is what actually moves revenue.

Why Your Brand Is Invisible to LLMs (And It's Not What You Think)

Veritas Links — Fri, 12 Jun 2026 23:41:57 +0000

This is not another AI slop-post, even if it's looks like on the first glance.

"SEO Is Not GEO." - say thousands post on LinkedIn everyday. But Nobody Actually Explains What That Means.

If you work in SEO — or manage it for clients — you've seen these posts. LinkedIn is full of them. Dozens every day, all saying roughly the same thing in roughly the same words. At some point they started to blur together, like they were all written from the same prompt. And there's a reason for that: most of them were. The people writing about GEO aren't actually doing GEO. They're recycling the same talking points without going one level deeper. So let me actually show you what's underneath.

Large language models don't crawl your website in real time. Instead, they build their understanding of your brand from a much wider and messier picture — everything that's been said about you across the internet.

They look at reviews, forum threads, articles, Reddit discussions, and industry publications. They notice how other sources categorize you — what market you belong to, who your competitors are AND how they are positioned against you, and what problem you AND they actually solve. They pay attention to how consistently your core message appears, and whether you're mentioned next to trusted names or completely unknown ones.

This is why a brand with average SEO but strong community presence and third-party coverage can easily beat a technically perfect website in AI answers. And why a brand with great SEO but almost no real-world footprint often becomes invisible.

The measurement problem nobody talks about
This is where things get interesting.

LLMs are stochastic by nature. Ask the same question twice and you can get noticeably different answers. One day your brand shows up, the next day it doesn't. Sentiment shifts. Competitor mentions change.

Most GEO tools (Profound, Otterly, Gauge, Scrunch, Brandwatch AI features, SEMrush/Ahrefs) try to solve this by running 10–50 prompts and averaging the results. The problem? With such a small sample, you're not measuring your real AI visibility — you're mostly measuring random noise.

At https://VeritasLinks.com/ we do it differently. Each report is built on 8 core metrics: AI Visibility, Share of Voice, Rankings, Competitors, Source Tracker, Mentions Explorer, Revenue Opportunity, and AI Perception.

For every metric we start with 10 base prompts. Every single response — no matter the original question — goes through a deep analyzer that extracts signals for all metrics. If a competitor is mentioned during a Mentions Explorer run, the system automatically takes it into account. It also analyzes the tone and context of every mention, both for your brand and your competitors.

If the answer isn't detailed enough or misses key information, the system doesn't stop — it asks follow-up questions, sometimes up to 10 levels deep. This allows us not just to spot patterns, but to genuinely pressure the model and dig deeper.

As a result, each metric can involve 10 initial prompts plus dozens of clarifying ones. This is how we get a much more stable and reliable picture.

What actually moves your AI visibility score

Based on analysis across hundreds of brands, the factors that consistently drive AI brand perception are:

Source diversity. If your brand is only mentioned on your own site and one or two directories, AI models have thin evidence to work with. They default to competitors who have broader third-party coverage.

Category consistency. If different sources describe what you do in different ways, AI systems get confused about where to place you. Inconsistent categorization leads to vague, hedged mentions — or none.

Recency of coverage. Models are regularly updated with new data. A brand that was active in industry conversations a year ago but has since gone quiet loses ground to competitors who are generating fresh mentions.

Competitor co-occurrence. If your competitors are mentioned alongside respected names in your category and you're not, the model learns to associate credibility with them, not you.

Trust signal density. Review platforms, case studies, industry mentions — these are the signals that tell a model your brand is credible enough to recommend.

A real example: the invisible service

One marketing agency ran a GEO analysis and discovered that their CRO (conversion rate optimization) service was almost completely absent from their AI visibility profile across all major models.

They hadn't removed it from their site. But they had stopped writing about it, stopped promoting it, stopped generating any new content or coverage around it.

The analysis estimated the revenue impact of that visibility gap at approximately $30,000 per month — based on market pricing for the service category and average contract values for agencies of that size.

When they checked their own internal numbers, their CRO revenue had dropped by roughly $34,000 per month over the same period.

The AI didn't predict this. It measured it. The visibility gap was already real — the GEO analysis just made it visible.

How to start measuring this yourself

The first step is understanding your current AI perception baseline — not what you think AI says about you, but what it actually, consistently says.

A few things worth checking manually:

Ask ChatGPT, Gemini, and Perplexity: "What are the best [your category] tools?" — Does your brand appear? In what position? How is it described?

Ask: "Tell me about [your brand name]" — Is the description accurate? Does your core value proposition survive? Are competitors injected into the answer?

Run the same prompts a week apart and compare. If the answers are significantly different, you're seeing stochastic noise — not real signal.

For a more systematic analysis across all major models, VeritasLinks runs 1000+ structured prompts per report and produces a stable VeritasScore — free for a single-model report, with paid plans covering all major LLMs. The free report is enough to see whether you have a problem worth solving.

The practical takeaway

GEO is not a replacement for SEO. You still need your pages to rank. But ranking is no longer the only moment of judgment.

AI answers are now the layer before the click. If your brand is invisible or misrepresented there, you're losing consideration before a potential customer ever reaches your website.

The good news: unlike traditional SEO, GEO visibility can shift relatively quickly. A few months of focused effort on third-party coverage, category consistency, and trust signal density can meaningfully change how AI models describe and recommend your brand.

The starting point is knowing where you stand right now.

If you found this useful, I write about GEO, AI brand perception, and the technical side of how language models form opinions about companies. Follow for more.