AI USER GUIDE
Navigating Chicardgo Data with AI
By Data Team | Published April 17, 2026 | Updated April 25, 2026
As artificial intelligence becomes a standard tool for analysis, many parents, educators, researchers, and policy makers use tools like ChatGPT, Claude, Grok, and Gemini to query high school ranking data. However, you may notice that different AI models sometimes provide varying or incorrect ECPI scores. This guide explains how different models process our data, the benefits and risks of each approach, and how to ensure you are utilizing the information accurately.
Understanding Generative AI
To use these tools effectively, it is essential to understand how Large Language Models (LLMs) function. Generative AI models are not databases; they do not "look up" facts in a traditional sense unless explicitly equipped with an integrated real-time search tool. Instead, they operate predictively.
Predictive Text Processing
Models process language by predicting the next most likely token based on their training data. This makes them excellent at formatting and summarizing, but prone to presenting statistically likely—yet factually incorrect—information.
Hallucinations
When an LLM encounters a gap in its training data (such as proprietary index scores), it may confidently fabricate plausible-sounding numbers to complete the pattern. This phenomenon is known as a "hallucination."
Knowledge Cutoffs
Many models are trained on historical snapshots of the internet. If a model lacks active web-browsing capabilities, its knowledge of Chicardgo ECPI rankings may be years out of date depending on its knowledge cutoff.
Why Do AI Models Provide Differing Results?
The Elite College Placement Index (ECPI) is proprietary. Our data tables are rendered dynamically on our website to ensure researchers and families always see the most up-to-date information. This creates a technical hurdle for AI:
The JavaScript Wall
Many AI web-scrapers only read "plain text" HTML. They cannot execute the JavaScript required to load our data tables. To some AIs, our ranking table simply looks empty.
Data Estimation
When an AI cannot read our table, it may attempt to estimate or calculate the ECPI score using outdated training data and publicly known names of elite high schools. This results in varying, unofficial numbers.
AI Crawler Management
Chicardgo School actively monitors and manages access from over a dozen distinct AI crawlers daily to protect proprietary datasets and ensure system stability.
By controlling these automated interactions, we maintain optimal site performance for human researchers while selectively allowing established AI assistants to access necessary context. We prioritize crawlers that respect our terms of service and provide direct value to our readership.
Current Access Protocols
Verified Search Engines
We permit access to established crawlers, such as Googlebot and BingBot, ensuring our public analysis reports and methodology pages remain accurately indexed and discoverable.
User-Driven AI Assistants
We allow specific user-agent crawlers, including ChatGPT-User, PerplexityBot, and Claude-SearchBot, to read dynamic content when explicitly prompted by a user conducting active research.
Aggressive Scraper Blocking
Our security infrastructure actively blocks non-compliant or unverified crawlers that attempt bulk scraping of our proprietary ECPI data tables without providing discernible user value.
How Different Models Behave
Based on our testing, here is how the leading AI platforms interact with Chicardgo's ECPI index:
Disclaimer: Chicardgo School does not endorse or mandate the use of any specific generative AI tools. The models listed below are evaluated solely to help our users navigate discrepancies in reported data. Users are responsible for evaluating the accuracy, privacy policies, and terms of service of any third-party AI platform.
Perplexity
Web-Crawling Approach
- Benefits: Uses advanced web crawlers capable of executing JavaScript, allowing it to read dynamically rendered data directly from our site for accurate citation.
- Risks: Responses can sometimes be overly brief, and it relies heavily on the live status of the website at the exact moment of the query.
Google Gemini
Search-Grounded Approach
- Benefits: Heavily grounded in search indexes. If it cannot access dynamic tables directly, it safely defaults to citing available aggregate data rather than estimating numbers.
- Risks: May refuse to provide specific row-by-row data if it determines the table is not fully indexed in its standard plain text crawler.
ChatGPT
Generative Approach
- Benefits: Highly conversational and excellent at formatting or synthesizing data if you provide the raw numbers directly in your prompt.
- Risks: If its browsing tool cannot execute JavaScript to read our live table, it may attempt to estimate ECPI scores based on internal training data, requiring careful user verification.
Claude (Anthropic)
Format-Oriented Approach
- Benefits: Extremely capable at building interactive charts, artifacts, and structuring complex analytical responses for researchers.
- Risks: Often prioritizes fulfilling the requested format. If it cannot read the live data, it may generate placeholder or rounded numbers to make an Artifact function.
Grok (xAI)
Real-Time Search & Social Integration Approach
- Benefits: Utilizes real-time web search tools to access dynamic content. When operating correctly, Grok retrieves data directly from live index tables and integrates current information.
- Risks: Output reliability depends heavily on the availability of its search connections and the factual accuracy of integrated X (formerly Twitter) data streams. If a direct search fails, it may rely on unverified social commentary or older training data to formulate a response.
Best Practices for Users
-
1
Always Verify on Chicardgo.com
The only indisputable source of truth for the Elite College Placement Index is our official website. Treat any AI-generated ranking as a starting point, not the final word.
-
2
Use Strict Prompting
If you must use an AI chatbot, restrict its behavior. Use prompts like: "Search the official Chicardgo School website. Do not estimate or calculate the ECPI score yourself. If you cannot read the exact data table, tell me."
-
3
Provide the Context Directly
If you are using ChatGPT or Claude to analyze school performance, highlight the rows you want from our website, copy the text, and paste it directly into your AI prompt rather than asking the AI to fetch the URL itself.
-
4
Iterative Prompting
Do not accept the first answer if it seems flawed. Refine your prompt by assigning a role, explicitly stating the desired format, and adding context. For instance, instruct the AI: "Act as an education data analyst. Compare the following specific data points..."
-
5
Maintain Data Privacy
When asking a GenAI tool to evaluate placements, do not input personally identifiable information (PII) or confidential student records. Assume that any prompt data may be utilized to train future iterations of the model.
-
6
Properly Cite Our Datasets
For journalists, researchers, and policy makers using our data in reports or training models, please use the standardized citation format when referencing our work. For example:
Chicardgo School (2026). Bay Area Private High School Elite College Placement Index Dataset [Dataset]. https://school.chicardgo.com/bayareaprivate/