Reading scan results
How to interpret the Results table, drill into a query, and view raw responses.
Last updated: 2026-05-05
TL;DR — The Results page lists every (prompt × model) execution from your scans. Each row shows whether your brand was mentioned and at what rank, which competitors appeared, what sources the model cited, and the status. Click any row to open the full raw response, parsed brand list, and metadata. Use Results when you need ground truth — the dashboard is aggregated, Results is row-level.
Why the Results page matters
The dashboard tells you what — your visibility is up 6 points, Acme is appearing alongside Canny 38% of the time. The Results page tells you why — exactly which prompts moved the needle, exactly what each model said, exactly which sources got cited. When something on the dashboard looks wrong, surprising, or worth investigating, Results is where you go to verify.
It's also where you'd:
- Audit the AI's actual wording (essential for PR, legal, or competitive intel)
- Build evidence for an internal report or stakeholder presentation
- Identify specific prompts that need refinement
- Spot brand mis-identification or alias gaps
- Pull source URLs for a citation-optimization project
How Results is structured
The Results page is one big filterable table. Each row = one (prompt × model) execution. Columns:
| Column | What it shows |
|---|---|
| Prompt | The exact text sent to the model |
| Model | Which AI engine answered (with provider icon) |
| Brand cell | Whether your brand was mentioned and its rank (e.g. "#3 of 8") |
| Competitors cell | Registered competitors found in the response, in order |
| Sources | Domain favicons of cited URLs (hover for full URL) |
| Status | Completed, No brands found, Failed (see Scan statuses) |
| Sentiment | Positive / Neutral / Negative classification of how your brand was discussed |
| Run date | When this execution completed |
| Country | Locale the scan was run for |
You can sort, filter, and search across all of these.
Drilling into a single row
Click any row to open the Result Details panel. You'll see:
- Full text response — the model's complete answer, exactly as returned. No editing or summarization.
- Parsed brand list — every brand the extraction engine found, in order of appearance, with rank and a confidence indicator.
- Source list — every URL the model cited (de-duplicated by domain slug — see SOV calculation standards).
- Sentiment breakdown — the overall sentiment classification plus the specific snippets that drove it.
- Token usage — input/output tokens and credits charged.
- Metadata — model version, timestamp, country, scan ID, retry count, original child-brand variants if your hierarchy consolidated mentions.
This panel is the single source of truth for that execution. If the dashboard shows something unexpected, this is where you reconcile.
Filtering Results
Above the table, you can filter by:
- Scan(s) — pick one or several specific scans
- Date range — defaults to last 30 days
- Country
- Model(s)
- Prompt type (Unaided, Aided, Comparison, Branded)
- Status (Completed only, include No brands found, include Failed)
- Brand mentioned (yes / no)
- Sentiment (positive / neutral / negative)
- Free-text search across the prompt and response
Filters compose. A common workflow: "show me all Unaided prompts on GPT-5 from the last 30 days where Acme was not mentioned" — that gives you the exact list of prompts to attack with content.
Worked example
After her weekly scan, Maya opens Results filtered to:
- Prompt type: Unaided
- Brand mentioned: No
- Date: last 7 days
The table shows 8 prompts where Acme was invisible. She clicks the top one: "best customer feedback tools for engineering teams".
The Result Details panel opens. The model recommended Canny, Linear (built-in feedback), and ProductBoard, citing two TechCrunch articles and a G2 listicle. Acme isn't in any of those sources.
Maya has her three-step plan:
- Pitch herself into the TechCrunch and G2 listicles (Action Center will likely pre-write this brief).
- Publish a "feedback for engineering teams" comparison page on her own site so the model has direct evidence.
- Add the prompt to her scan-set permanently to track recovery.
That entire investigation took 3 minutes from the dashboard observation to the action plan.
How dashboard metrics map to Results
| Dashboard metric | Computed from Results rows where… |
|---|---|
| AI Score | all completed unaided rows, weighted by rank quality, appearance rate and stability; non-mentions count as zero |
| SOV | brand mention counts vs. registered competitor mention counts |
| Top Brands | aggregated brand mentions across rows |
| Co-occurrence | brand pairs appearing in the same row |
| Sources | URLs from the source column, deduped by domain |
| Perception | sentiment + drivers extracted from response text |
If a dashboard number looks wrong, you can always reproduce it by filtering Results to the same scope and counting.
Edge cases & gotchas
- Aided/Branded prompts inflate "mentions" — when reading raw Results, remember that aided prompts ask about your brand by name, so mentions are guaranteed. Filter to Unaided for organic signal.
- Source dedup: Results details show all URLs returned; the dashboard dedupes by URL slug. Different surfaces, same underlying data.
- Variant metadata: if your brand consolidates child brands under a parent, the parsed brand list shows the original variant for transparency.
- No brands found rows still get a detail view — useful for understanding why the model didn't recommend you.
- Failed rows show whatever partial data was captured (sometimes a partial response with extraction skipped).
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Brand cell empty but you read the response and saw your brand | Alias missing from extraction | Add the variant as an alias on your brand |
| Competitor cell missing a competitor you saw | Competitor not registered | Register the competitor (see Manage competitors) |
| Source list shorter than what you see in the response text | Extracted vs. hallucinated URL filtering | Pranas validates URLs; hallucinated ones are stripped |
| Filters don't return what you expected | Stacked filters too narrow | Reset filters one at a time |
Related articles
Still need help?
Email contact@pranas.co with the scan ID and the row(s) you're investigating.