Ai Marketing4 min read

AI Competitive Intelligence: Stop “Confident Hallucinations” with Tess

January 30, 2026By Sandbox Media

TL;DR: When you ask generic AI to research a competitor’s pricing or features, it confidently gives you made-up data. In this episode, we show how Tess, our AI Competitive Intelligence Officer, uses Truth Filters and live browser verification to turn hallucinations into actionable strategy.

Here is a terrifying thought for any business leader using ChatGPT for strategy.

If you ask an AI model, “What features are in my competitor’s Premium Plan?”, it will give you an answer. It will look polished. It will include bullet points. It might even include a price.

And there is a high probability that the information is completely made up.

This phenomenon is called “Confident Hallucination.” Standard Large Language Models (LLMs) are Prediction Engines, not Search Engines. Their goal is to predict the next plausible word in a sentence, not to verify facts. If the AI doesn’t know the answer, it will often invent a “plausible” number just to complete the pattern.

If you make a strategic decision based on a hallucinated price point, you aren’t using AI. You are gambling.

At Sandbox Media, we don’t let AI guess. We use Tess, our designated AI Competitive Intelligence Officer. Tess runs on a strict architecture of Truth Filters.

Building “Truth Filters”

To turn an AI from a “Creative Writer” (who guesses) into a “Research Analyst” (who verifies), you need two specific injections: an Instruction Set and a Brand Blueprint.

Injection 1: The Instruction Set (Journalistic Standards)

First, we enforce “Epistemic Humility.” We have to teach the bot that “I don’t know” is a better answer than a lie.

We run Tess on over 1,000 lines of custom instructions that act as a strict editorial code. The key constraints:

The Browser Constraint: “You are FORBIDDEN from relying on your internal training data for numbers, dates, or prices. You MUST use the live Browser Tool to verify current data.”
The Citation Rule: “Every claim must be backed by a specific, clickable URL source. No source = No claim.”
The Null Option: “If you cannot find live verification, you must state ‘DATA UNAVAILABLE.’ Do not estimate.”

These constraints force the AI to stop predicting and start investigating.

Injection 2: The Brand Blueprint (The Strategic Radar)

Second, we give the AI context. A generic research assistant Googles everything. A strategic analyst looks for Leverage.

We utilize the Competitor Radar and Strategic Positioning sections of the Brand Blueprint to tell the AI exactly how we compete:

Who are we fighting? (e.g., “We are fighting Salesforce”)
What is our angle? (e.g., “We win on Total Cost of Ownership / Hidden Fees”)

Case Study: FlowState CRM

Hypothetical client: “FlowState CRM,” positioning itself as a transparent alternative to big enterprise software.

The Generic AI (The Hallucinator)

Input: “Compare FlowState’s pricing to Salesforce.”

Output: “Salesforce offers a starter plan at $25/month. It includes basic contact management…”

Verdict: Dangerous. This data might be from 2021. It makes Salesforce look cheap. It ignores contract nuances entirely.

The Tess Approach (The Analyst)

Tess is given the Instruction Set (Must Browse) and the Blueprint (Look for Hidden Fees). She scrapes the current pricing page and the Terms of Service PDF.

Input: “Analyze the Total Cost of Ownership gap.”

Output: “WARNING: While the Salesforce base price is listed at $25, I found a clause in their current Terms of Service [Source Link] regarding ‘API Call Limits.’ Based on your Blueprint, your target customer (Mid-Sized Teams) will hit this limit in Month 3, forcing an upgrade to the $150/user tier. Strategy: Do not compete on the $25 price. Build a comparison calculator showing the ‘API Trap’ that raises their cost by 600%.”

Verdict: The Generic AI gave a number. The Trained AI gave a Winning Strategy.

Stop Guessing. Start Verifying.

In the age of AI, information is cheap. Verified Intelligence is expensive.

You cannot afford to build your business strategy on hallucinations. You need a system that values truth over plausibility.

Build the Instruction Set: Force citations.
Build the Blueprint: Define your strategic angle.
Run the Browser Tool: Verify everything.

Ready to stop gambling and start building real competitive intelligence? Start with our Sandbots Consultation or build your Competitive Radar with our AI Branding & Guardrails Consult.