Overview

AI-powered search tools like ChatGPT, Perplexity, and Gemini are becoming default starting points for customer discovery. But they work differently from traditional search engines—retrieving, ranking, and citing web content in unique ways. This guide answers some of the most advanced questions we hear from customers navigating AI-driven visibility and explains what Scrunch is doing to help.

Frequently Asked Questions

1. How many prompts do I need to track to understand product mentions and sentiment within a topic?

There’s no magic number, but the key is representative sampling. AI models may surface very different results depending on:

Prompt phrasing
Model used (ChatGPT vs. Perplexity, etc.)
Recent memory or session history (in ChatGPT)
Web content indexed or retrievable in real-time

Scrunch recommends focusing on 15–25 varied prompts per topic. Our platform re-runs each prompt regularly across supported models, giving you hundreds of real-world data points to analyze citation frequency, position, and sentiment over time.

Tip: Use Scrunch’s Key Topics and Prompts tab to group and analyze related queries.

2. How much does context or memory change a chatbot’s response?

In ChatGPT, quite a bit. OpenAI confirms that custom instructions, past conversations, and saved memories can alter both the search terms used and the tone or framing of the output (source). Even with identical pages cited, the result might differ depending on whether a user is asking from a blank slate or mid-conversation.

Scrunch helps by simulating neutral, cold-start prompts so that your visibility isn’t distorted by individual user context.

3. Why do the pages cited in RAG (Retrieval-Augmented Generation) diverge from the top-ranked search results?

This is expected. Most AI search systems retrieve pages using traditional ranking (e.g., via Bing), then re-rank or re-select based on the relevance of the snippet to the question - not the page's SEO rank.

So:

A page ranked #5 in Bing might be cited first by ChatGPT.
Re-ranking is usually handled by a separate model trained to evaluate snippet quality and coverage.

Scrunch gives you insight into both the 'under the hood' results, as well as the reranked pages being cited within the answer.

4. Why do some citations from a brand recommend their own competitors?

AI often cites third-party sources with product lists, like blog posts or “best tools” roundups. The recommendation can differ from the article’s original intent. For example:

A HubSpot listicle might mention 10 CRMs.
If a prompt asks “what’s the best CRM for freelancers?” ChatGPT may cite that page but summarize a competitor’s product as the top option.

This is normal behavior - AI selectively pulls a subset of the information on the page. That’s why presence in the source content matters, even if you’re not the site owner.

Solution: Earn placement on high-authority third-party content. Scrunch’s Sources tab shows what URLs are driving citations.

5. How different is the ChatGPT API vs. the ChatGPT App experience?

They often diverge:

The App supports memory, custom instructions, and personal history.
The API (and non-logged-in Web app) are “cold start” and more consistent across users.
Citation behavior, snippet usage, and model version may differ slightly across environments.

6. My brand is not getting significant referral traffic from AI sources according to Google Analytics. Is that the full picture?

No - it’s likely just the tip of the iceberg.

Referral traffic captured in Google Analytics (e.g. via ChatGPT or Perplexity referrers) only includes clicks from links inside the AI interface. But AI often drives indirect traffic, such as:

Users seeing your brand in an AI result and then Googling it
Direct visits from users who type in your domain
Branded search lifts

Scrunch is actively developing methods to approximate these broader impacts. If you’re interested in advanced tracking or attribution modeling, keep up with our beta launches here.