AI & Technical question

You’re part of the Google Search web spam team. How would you detect duplicate websites?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Tests technical product thinking for a search-quality/spam problem, requiring concrete detection signals and a sense of tradeoffs in enforcement.

Define duplicate types: exact mirror content (scraped sites), near-duplicate with minor spun text, and templated doorway pages with different URLs pointing to the same destination intent.
Propose detection signals: content similarity hashing (shingling/simhash) to catch near-duplicates cheaply at scale, plus structural signals like identical site architecture or shared hosting/registration patterns.
Add behavioral signals: near-identical backlink profiles or coordinated link patterns between suspected duplicate domains, which are hard to fake convincingly.
Design action tiers: for exact duplicates, canonicalize and rank only the original source; for spam-intent duplicates (doorway pages), demote or de-index rather than just merge.
Address the tradeoff: aggressive similarity thresholds risk false positives on legitimately similar content (e.g. syndicated news), so combine multiple signals rather than relying on text similarity alone.

Names concrete, real detection techniques (shingling/simhash, backlink pattern analysis) instead of a vague 'use an algorithm to detect duplicates'.
Distinguishes benign duplication (syndicated content) from malicious spam duplication (doorway pages), which need different responses, not the same penalty.
Proposes combining multiple independent signals to reduce false positives, a key defensive-systems design principle for adversarial spam problems.
Ties the solution to search quality's actual goal, showing understanding this protects both user trust and ranking fairness for original content creators.

Proposing text-similarity detection alone without addressing that spammers actively evade naive similarity checks.
Treating all duplicate content the same way instead of distinguishing benign syndication from malicious spam intent.

How would you handle legitimate syndicated content, like a press release published on many news sites?
How would spammers try to evade your detection system, and how would you adapt?