Understanding match/no-match results and AI reasoning

Every listing analyzed by FilterWise comes with a detailed breakdown of the AI's decision. Understanding how to read this analysis helps you refine your instructions and trust the results.

Where to See Results

Click any listing in the Listings page to open the detail panel. The AI analysis section shows the complete breakdown.

The Conditions Table

The conditions table is the core of the analysis. It lists every criterion from your filter instructions as a separate row:

Column	What It Shows
Criterion	The specific requirement from your instructions
Value	What the AI extracted from the listing for this criterion
Match	✅ Yes, ❌ No, or ❓ Uncertain

Example

For instructions like:

Forward if:
- Price under €400,000
- At least 3 rooms
- Not ground floor
- Quiet street

The conditions table might show:

Criterion	Value	Match
Price under €400,000	€350,000	✅
At least 3 rooms	4 rooms	✅
Not ground floor	2nd floor	✅
Quiet street	"Located on a calm residential street"	✅

Overall verdict: Match — all criteria satisfied.

How the AI Determines Match/No-Match

Step 1: Criterion Parsing

The AI breaks your instructions into individual, evaluable criteria. Each becomes a row in the conditions table.

Step 2: Value Extraction

For each criterion, the AI scans the listing data (title, description, specifications, scanned page content) to find the relevant value.

Step 3: Evaluation

The AI compares the extracted value against your requirement:

• Numeric comparisons: Price < €400,000, rooms ≥ 3, floor > 0
• Boolean checks: Has balcony (yes/no), needs renovation (yes/no)
• Text matching: Location includes "Paris 11th", description mentions "calme"
• Calculations: price/m² < €7,000, cash flow > €200

Step 4: Overall Verdict

The verdict depends on your instructions's logical structure:

• AND logic: ALL criteria must match → overall Match
• OR logic: ANY group must match → overall Match
• Mixed: Evaluated according to the logic groups you defined

Understanding "Uncertain" Results

A criterion is marked uncertain (❓) when the listing doesn't contain enough information to evaluate it. Common examples:

• Floor level not mentioned → "Not ground floor" is uncertain
• No noise-related keywords → "Quiet street" is uncertain
• Surface area not stated → "Price per m² under €7,000" is uncertain

The "include when unsure" setting controls whether uncertain criteria are treated as matches or failures. See Understanding include when unsure for details.

Reading the AI's Reasoning

Below the conditions table, the AI provides a text summary of its reasoning. This is especially useful for understanding:

• Why a value was extracted: "The listing mentions '3 pièces' which translates to 3 rooms"
• Why something is uncertain: "The listing does not mention floor level or étage"
• How calculations were performed: "Price (€350,000) ÷ surface (55m²) = €6,364/m², which is below the €7,000 threshold"
• Contextual interpretation: "The address includes 'impasse' which typically indicates a quiet dead-end street"

Using Results to Improve Your Instructions

The analysis results are your best tool for refining your filter instructions:

Too Many False Negatives (Good listings rejected)

Symptom: Listings you would have liked are showing "No Match."

Diagnosis: Check which criterion caused the rejection. Common fixes:

• Threshold too strict: Raise the price limit, lower the room requirement
• Too many uncertain criteria: Enable "include when unsure" or remove criteria that listings rarely mention
• Location too narrow: Expand the geographic area

Too Many False Positives (Bad listings matched)

Symptom: You're being forwarded listings you don't want.

Diagnosis: Check which criteria are consistently satisfied too easily. Common fixes:

• Threshold too loose: Lower the price limit, increase minimum rooms
• Missing criteria: Add additional requirements you care about
• Vague wording: Be more specific about what you mean

Calculations in Results

When your instructions includes calculations (price/m², cash flow, yield), the conditions table shows:

• The formula used
• The input values extracted from the listing
• The computed result
• Whether it passes your threshold

This is invaluable for verifying that the AI computed correctly. If a calculation seems wrong, check whether the AI extracted the right input values.

Model Information

Each analysis shows which AI model was used. FilterWise uses state-of-the-art language models for listing analysis. Different models may occasionally produce slightly different interpretations of the same listing — this is normal.

Re-Analyzing a Listing

If you've updated your instructions and want to see how an existing listing would be evaluated with the new criteria:

1. Open the listing detail panel
2. Click Re-analyze
3. Confirm the action (it counts toward your daily limit)
4. The analysis is re-run with your current instructions and fresh scanned data

---

Related articles: Writing effective instructions · Understanding include when unsure · Browsing your listings