Logo
Back to Blog
Security August 9, 2026 8 min read

Google Dorking for Bug Bounty Recon — A Practical First Pass

O

OmniWebKit Team

Security Tooling

Share:
Article Cover Image

You pick up a new programme and immediately start firing scanners at it. Two hours later you have noise, rate limits, and nothing a triage team wants. Search-based reconnaissance costs nothing, touches no infrastructure, and routinely finds things the scanners never see.

Google Dorking for Bug Bounty — Where It Fits

It is a first filter, not a methodology. Thirty to sixty minutes at the start of a programme, then move on.

The appeal is that nothing you do reaches the target. Your queries go to Google. No traffic, no logs on their side, no rate limits, no chance of tripping something during a maintenance window.

What it finds well: indexed files nobody meant to publish, forgotten subdomains, old marketing sites, documents with useful internal detail. What it finds badly: anything current, anything unlinked, anything behind authentication.

New to the technique itself? How dorking works covers the mechanics. This piece assumes you have that and are working an authorised programme.

Scope Verification Comes Before Any Query

Read the scope document first, every session, not once when you enrolled. Scopes change mid-programme and the change is rarely announced loudly.

Three things to establish before you type anything:

  • Which domains are in. Wildcards vary — some programmes include all subdomains, many do not.
  • Which are explicitly out. Out-of-scope lists exist because those assets belong to third parties or are mid-migration.
  • What the rules say about reconnaissance. Most are silent, which means normal searching is fine. A few are explicit.

An asset that obviously belongs to the organisation but is not listed is out of scope. That is not a technicality — it is frequently a supplier's infrastructure, and testing it creates a problem for everyone. We cover the wider question in authorisation and the law.

A Passive Reconnaissance Sequence That Works

Broad first, then narrow. Reversing this wastes the hour.

Start with a plain site: search on the primary domain to get a feel for structure and volume. Then invert it — site:target.com -inurl:www — to surface content on other hosts. That single query finds more forgotten infrastructure than anything else in this list.

Next, formats. Documents, spreadsheets and presentation decks carry internal detail that HTML pages do not: staff names, system names, internal URLs, occasionally architecture diagrams.

Then paths. Directory listings, repository folders, diagnostic pages. These are the classic exposures and they are still there because nothing about deploying software got easier.

Finally, error output. Indexed stack traces reveal framework versions and file paths, and they also tell you a route is broken.

The specific query patterns for each stage sit in the dork categories worth trying first, so we will not duplicate them here.

Why Search Recon Finds What Enumeration Misses

The two techniques sample completely different sources.

Subdomain enumeration works from DNS records and certificate transparency logs. Search works from what a crawler happened to follow. A host with no certificate and no DNS wildcard can still be indexed because someone linked to it from a forum post in 2019.

That is why the overlap between the two is smaller than people assume. Running both is not redundant effort — we have found live staging environments through search that no enumeration tool surfaced.

Worth repeating on Bing. Its index differs meaningfully from Google's, and the same query against both regularly returns non-overlapping results. The recon query generator switches engines from one dropdown, which makes the comparison quick.

Reading Results Without Wasting Triage Time

Most of what these queries return is noise, and reports built on noise cost you standing.

Three things that are not findings on their own: an indexed login page, a deliberately published document, and a subdomain that simply exists. Triage teams see these constantly and they close them fast.

What is worth writing up: a config file with real values, a database export, a repository directory, a diagnostic page listing your environment. The test is whether it hands over something the organisation would not choose to publish.

Verify before reporting. Low-traffic paths go unrecrawled for months, so a result can point at something already removed. A report resting on a stale index entry damages credibility more than staying quiet would.

If you find credentials, do not use them. Confirming the file exists is the finding. Authenticating with its contents is a different act entirely, and the consequences are not comparable.

What This Approach Cannot Do

Search only sees what was crawled, which is a smaller slice than people expect.

An unlinked file can be fully reachable and completely invisible to every query you write. A clean search pass tells you nothing about the security of anything — it tells you nothing was indexed.

It is also blind to logic flaws, authentication problems, access control failures and everything else that constitutes most real bug bounty work. Treat it as the cheap thirty minutes before the real work starts.

Understanding it from the other side helps more than it sounds. Our defender's version of this workflow pairs each query with its remediation, which teaches you quickly what a genuine finding looks like versus what a site owner would shrug at.

Making Search Recon a Habit Worth Keeping

Keep five queries you have proved return something, not a list of a thousand. Long lists feel thorough and are mostly dead entries duplicating each other.

Run them manually. Automating bulk queries breaks Google's terms, gets you rate-limited within minutes, and looks materially different if anyone ever reviews what you did.

Check operator status occasionally too — several commands still circulating in recon guides stopped working years ago and fail silently. The operator status reference saves testing each one yourself.

Before you start, confirm the target resolves and belongs to who you think. A DNS record lookup and a quick check of what the host is running with the stack detector takes two minutes and occasionally changes the plan entirely.

Frequently Asked Questions

Does search reconnaissance count as testing under a programme scope?

+
Searching a public index touches no infrastructure, so most programmes do not treat it as testing. Acting on a result does. Read the specific programme rules rather than generalising, because a few are explicit about reconnaissance methods.

A subdomain turned up that is not listed in scope. Can I test it?

+
No. An unlisted asset is out of scope until the programme says otherwise, however obviously it belongs to the same organisation. Report it as an asset discovery if the programme accepts those, then wait.

Why does search recon find things that subdomain enumeration misses?

+
They sample different sources. Enumeration works from DNS and certificate data; search works from what got crawled. A host with no certificate and no DNS wildcard can still be indexed because someone linked to it once.

How stale are these results likely to be?

+
Low-traffic paths can go unrecrawled for months, so a finding may reference something already removed. Always verify a result resolves before writing it up. A report based on a stale cache entry damages your credibility.

Is it worth running the same dorks on Bing?

+
Yes, and it is the highest-value duplicate effort available. Bing indexes a meaningfully different set of pages, so the same query against both regularly returns non-overlapping findings.

Do programmes accept findings that came purely from search?

+
An exposed file is a valid finding regardless of how you found it. What triage teams reject is a report that presents an indexed login page or a public document as a vulnerability. The bar is impact, not discovery method.

Should I automate these queries?

+
Manual queries stay well within normal use. Automated bulk querying breaks Google's terms, gets your address rate-limited quickly, and puts you in a different category if anyone reviews what happened. We do not recommend it.

How much time is reasonable for a search recon pass?

+
Thirty to sixty minutes on a first look. Past that, returns drop sharply and other techniques pay better. Search is a cheap first filter, not a complete methodology.

What if I find credentials in an indexed file?

+
Do not use them. Confirming an exposed file exists is the finding; authenticating with what it contains is a separate act with entirely different consequences. Screenshot the exposure, redact the values, and report.

Does any of this need a VPN?

+
No, and it can complicate matters by changing which jurisdiction your traffic appears to come from. Ordinary searching against a programme you are enrolled in needs no concealment.

Tags

#Security#Bug Bounty#Recon#OSINT