Choose AI models for a scan

Mix providers (OpenAI, Google, Anthropic, Meta, etc.) — or use the Google AI Overview default.

Last updated: 2026-05-05

TL;DR — Pick 2–4 models per scan to balance signal and cost. Google AI Overview is the default for new scans. Mix providers (Google + OpenAI + Anthropic) to detect cross-model differences. Premium reasoning models cost more credits per query — locked tiers show a 🔒 with the required plan.

Why model selection matters

Different AI engines answer the same question differently. ChatGPT might recommend Stripe; Perplexity might recommend Stripe + Adyen + Mollie with citations; Claude might give a more cautious "depends on your needs" answer. Your prospects use different tools, so your visibility data has to span them.

Choosing the right mix of models is the single biggest lever on:

• Insight quality — provider variety reveals divergence (e.g. you're #2 on ChatGPT but invisible on Gemini)
• Credit cost — premium reasoning models cost 5–10× more per query than fast/cheap ones
• Scan speed — heavy models batch smaller and take longer
• Plan tier alignment — some models are gated on Growth/Pro

The supported model families

Pranas integrates 17+ models through OpenRouter and Google's APIs. The current roster includes:

The full active roster is shown in the model picker; models that fail health checks are temporarily disabled and hidden.

The default: Google AI Overview

When you create a fresh scan, Google AI Overview (Gemini 3.5 Flash-Lite with live Google Search) is pre-selected. This is intentional — Google AI Overviews now appear above traditional search results for billions of queries, making it arguably the highest-leverage AI surface to monitor today. If you only run one model, run this one.

You can change the default per-scan; the system remembers your last selection.

Step-by-step: select models for a scan

1. Open Run AI Scan (Dashboard topbar or Results → New Scan).
2. In the dialog, scroll to the Models section.
3. Click any model card to toggle it on/off.
4. Locked models show a 🔒 lock icon and the minimum plan required (e.g. "Growth"). Click for an upgrade prompt.
5. As you toggle, the estimated credit total at the bottom updates live: (prompts × selected models) × per-model cost.
6. When happy, click Run Scan.

Recommended starter combos

For most weekly scans, 3 models is the sweet spot: enough variety to catch divergence, low enough cost to scan often.

Premium model gating

Some advanced models are gated by plan tier:

• Free / Trial / Starter → fast and mid-tier models
• Growth → most models, including some premium
• Pro → all models, including top-tier reasoning (ChatGPT Pro, Gemini 3.1 Pro, Claude)

The exact gating shows in the model picker — locked models display the required plan inline.

Worked example

Maya runs a weekly 25-prompt scan. She wants meaningful signal without burning credits:

1. Default scan: Gemini + GPT-6 Sol + Claude Haiku 4.5 (3 models, ~75 executions, ~$X).
2. Monthly deep audit: she adds Gemini 3.1 Pro and ChatGPT Pro (5 models, ~125 executions, ~3× the cost) for one scan per month to catch what the lighter models miss.
3. Quarterly cross-model comparison: she runs every model on a focused 10-prompt set to map cross-model perception drift.

Three different scans, three different cost/insight tradeoffs.

Edge cases & gotchas

• More models ≠ proportionally more insight — there's diminishing return after 4–5 models.
• Model health checks automatically disable models that are failing. If a model you expected to see is missing, it's likely temporarily unhealthy.
• DeepSeek R1 and similar reasoning-stream models require special parsing; if their responses are unparseable they're marked as failed (not retried indefinitely).
• Adding a model mid-scan: not supported. Launch a new scan with the additional model on the same prompts if you want the comparison.
• Premium model billing: premium models cost more credits per query than lighter models. The credit estimator always shows the all-in cost before you launch.

Troubleshooting

Related articles

• Run your first AI Scan
• Credits per scan
• Scan queue and retries
• AI Models
• Plans overview

Still need help?

Email contact@pranas.co if a specific model isn't behaving as expected — include the scan ID and model name.