Why I built it
I got tired of the usual workflow:
- Exporting GSC data into spreadsheets or Looker Studio
- Merging with crawl data or running manual audits
- Writing prompts by hand for LLMs
- Fighting the API’s limits, sampling, and UI quirks
- Trying to keep a simple story straight across thousands of keywords
So I built something that:
- Uses only your own verified Search Console data (privacy-first)
- Runs a structured audit pipeline using Gemini (chosen for its massive context window)
- Makes it easy to find the “needle in a haystack” keywords with advanced filtering
- Classifies search intent using a trained Sentence Transformers model
- Parses massive HTML pages (up to 250K tokens)
- Surfaces clear, actionable content insights without fluff
- Focuses on modern SEO best practices (search intent, EEAT, helpful content)
What’s under the hood
Stack: React (frontend), Django (backend), Celery (async), Google OAuth (auth)
Metrics engine: 6-part dashboard (SEO, EEAT, Content, Intent, Gaps, Speed)
LLM pipeline: Config-driven audit engine using Gemini (with Google Search grounding for reputation checks)
Search intent: In-house Sentence Transformers classifier (retrainable via user feedback)
Affiliate-friendly: 30% recurring commission + first-month discount code
Technical challenges
Gemini’s JSON output can get buggy with large HTML input (250K+ tokens). We found it far more reliable to return structured markdown, which we then parse into clean JSON on our side.
“Just a ChatGPT wrapper”?
Nope (we use Gemini ), but more seriously:
While LLMs power some audits, the audit pipeline is model-agnostic — we could swap in Claude, OpenAI, or an open-source model. Prompts are system-tuned to deliver actionable, critical feedback in a structured format that mirrors how I do real SEO audits.
Sure, you could copy/paste your HTML into AI Studio, craft a prompt, and manually parse the result. But:
- You'll hit token limits
- You'll get inconsistent outputs
- You’ll have to do that every time
Or you could just use QueryBurst and get the insight without the overhead.
Privacy-first by design
You can only see data from properties you’ve verified in your own GSC account. We don’t build a global keyword database, don’t allow competitor lookups, and don’t store user passwords.
Data is stored for reporting, but deletions are real (not just “hidden” flags). And again — your data is your data. Nobody else sees it.
Search Intent (and plans to open source)
We classify keyword intent using a fine-tuned Sentence Transformers model. If a classification is off, users can correct it — and we plan to retrain periodically on this anonymized feedback.
We’re also considering open sourcing the training microservice (Dockerized) that handles this, to give others a fast way to train their own classifier on internal data.
Not open source, but fully transparent
You do need to connect your GSC account to use it (since that's where your data lives). But you can watch a 1-hour walkthrough video before signing up — see exactly how it works and what to expect at https://queryburst.com
There’s a free version (with limited data), and paid plans start at $45/month.
Would love your feedback
Especially from devs or SEOs frustrated with the way GSC presents data — or anyone curious about building LLM-based tools that go beyond wrapping an API.
Happy to answer any technical, SEO, or LLM questions in the comments.