Different Ways to Use Your AI Data Pack

You ran the analysis. You read the six insights, fixed the obvious things, and rewrote the two pages that were clearly letting you down. Now you want to go further, and the report has told you everything it is going to tell you.

That is what the AI Data Pack is for. It comes with every analysis, it is one file, and it holds the evidence the report was built from. The report is the conclusion. The pack is the working.

What is actually in the file

It is a single markdown file, around half a megabyte for a typical run, and it contains five things:

• A briefing. A description of your business, how the data was collected, your headline results, and the six insights already identified. It is written for an AI assistant to read first.
• The mention map. One row per question, one column per AI model, showing which models named you and which did not.
• What the models searched for. The actual search queries the assistants ran while answering.
• The source index. Every site the models cited while answering, ranked by how often, and labelled with which model cited it and whether it belongs to a competitor.
• All 125 responses, verbatim. Every answer from every model, in full.

Two things about it are worth knowing before you start. It is built directly from your stored results with no AI involved, so downloading it twice gives you the identical file. And it contains no personal data.

Why the raw responses matter

The report answers "how visible am I, and what should I do about it." The pack answers the questions underneath that: which exact phrasings lose you the answer, which sites the models keep going back to, and how a competitor gets described when you are not in the room.

That last one is the most useful and the least comfortable. Somewhere in those 125 responses, an assistant explains to a buyer why a competitor is a good choice. You get to read the pitch that beat you, word for word.

If you use an agentic coding tool

If you work with something like Claude Code, Grok Build, Codex, Gemini CLI or Cursor, this is where the pack earns its keep, because those tools can read your website's source and the pack at the same time. That combination is the difference between advice and a change.

Put the file somewhere the tool can see it, alongside your site, then work through it:

• Find the gaps, then fill them. Ask it to list every question in the mention map where no model named you, check whether your site has a page that answers each one, and write the ones that are missing. It can compare the question to your actual pages rather than guessing.
• Build the FAQ from the informational questions. The pack marks which questions are informational. Those are the ones people ask while they are still learning, and they are usually the easiest to answer well.
• Work the source index. Those are the sites the models already trust in your market. Ask the tool to go down the list and tell you, honestly, which ones you could plausibly appear on and how.
• Read the competitor's pitch. Point it at a competitor that appears often, ask it to collect every passage describing them, and then ask what those passages have in common that your own pages do not say.

If you use a free or low-cost assistant

You do not need a coding tool. The pack was written to be read by any assistant, and the free tiers handle it well if you set them up once rather than pasting into a chat window.

• Use a notebook tool as a searchable library. Google's NotebookLM is the clearest example: upload the pack as a source and it becomes something you can question repeatedly, with answers pointing back to the part of the file they came from. This suits a large file better than a chat window does, because you are querying it rather than trying to hold it all in one conversation.
• Attach it to a project. If your assistant has projects or workspaces, ChatGPT and Claude both do, add the pack once as a project file. Every conversation you start in that project can then draw on it, and you stop re-explaining your situation.
• Split it if you have to. The bulk of the file is the prompt-by-prompt responses. The analysis lives in the first few sections. If a tool struggles with the whole thing, load the briefing, mention map and source index first, which is exactly what the file itself tells the assistant to do.

The one thing we would avoid is pasting the whole file into an ordinary chat. It is large, and most of what makes it useful is the ability to go back to it.

Ask better questions of it

The pack ships with a short list of suggested tasks, written for your specific results, so start there rather than from a blank prompt. They are worth reading even if you do not use them, because they show the kind of question the data can actually answer.

Beyond that list, the questions that tend to pay off are comparative rather than general. "How do I improve my AI visibility" is too broad for any dataset to answer well. "Compare how the models describe the top competitor with how they describe us, quoting both" is a question this file can genuinely settle.

One safety note

The responses in the pack came from external AI models, and they are quoted verbatim. The file tells any assistant reading it to treat that text strictly as data to analyse and never as instructions, which is the right default for any content that arrives from the open web.

That instruction is in the file. It is worth knowing it is there, and worth keeping it there if you ever cut the file down.

Snapshots, not live readings

The pack records what the models said on the day you ran the analysis. AI answers drift as models change and as the web underneath them changes, so treat it as a dated measurement rather than a permanent description of your market. The file carries its own date for that reason.

When you have made real changes and want to know whether they moved anything, a re-run puts the same questions to the same models again, which is what makes the two measurements comparable.

Key takeaways

• The report is the conclusion, and the AI Data Pack is the evidence it was built from.
• It is one markdown file, built without AI, identical every time you download it, and free of personal data.
• With an agentic tool that can also see your site, the pack turns into changes rather than advice.
• With a free assistant, set it up once as a source or a project file instead of pasting it into a chat.
• Ask comparative questions. The file is much better at "how do they describe them versus us" than at "what should I do".