Skip to content

Topic 3 min

Reranking

Being retrieved is the audition; reranking is the casting call: authority, freshness and answer-fit decide who makes the answer.

Authority signalsFreshness weightingAnswer-fit scoring

Why it mattersPlenty of retrieved passages never get cited; the rerank is where most are cut.

01

Definition & Foundation

What it is, in plain words

Being retrieved is the audition; reranking is the casting call. After retrieval assembles a shortlist of semantically relevant chunks, a second scoring pass re-orders and prunes it, and this is where most retrieved passages die. The rerank asks harder questions than "is this relevant?": Can this source be trusted? Is it current? Does it answer the exact question, completely?

The signals converge across studies of AI Overview and answer-engine citations: authority (recognizable sources, named authors, strong E-E-A-T; ~96% of AI Overview citations show strong E-E-A-T signals), freshness (recently published or updated content is disproportionately cited; roughly 44% of AI Overview citations come from content published within the last year), and answer-fit (passages that fully answer the question are ~4.2× more likely to be cited than partial answers). Relevance gets you retrieved; these three get you kept.

02

The Three Rerank Signals

What the casting call actually scores

can you be trusted?

Authority

Named authors with credentials, a recognized brand entity, citations from other sources, consistent facts across the web. Machine-parseable trust (see E-E-A-T for Machines), not self-declared expertise.

is it current?

Freshness

Recently published or genuinely updated content gets a rerank boost, especially on topics where answers change. A visible, honest last-updated date plus actually-refreshed content beats a fake timestamp; see Freshness & Cadence.

is it complete?

Answer-fit

Does the passage answer the exact question, fully, on its own? Semantic completeness is the strongest of the three in citation studies: a passage scoring high on completeness is ~4.2× more likely to be cited. Partial answers audition well and get cut here.

03

Myths vs Reality

Common misreadings, corrected

Myth"We got retrieved; the citation will follow."
RealityMost retrieved passages never survive the rerank; the shortlist is long and the answer cites a handful. Surfacing in an engine's source list without ever being cited is the signature of a rerank loss, not a retrieval win.
Myth"Authority means domain rating."
RealityClassic link metrics correlate weakly with answer-engine citations. The rerank reads authority as parseable signals: named authors, entity recognition, being cited by other retrieved sources, factual consistency, which small sites can win and big domains can fail.
04

Putting It to Work

Strengthen the three signals on the pages that matter

You can't see the reranker's scores, but you can systematically raise the three inputs it reads. Work page by page, signal by signal:

The signal-strengthening playbook

1

Score your target pages on the three signals

For each page that should earn citations: does it have a named, credentialed author? A real last-updated date under 12 months? A passage that completely answers the target question? A simple red/yellow/green per signal reveals the weak one.

2

Fix authority with attribution, not adjectives

Add a named author with a bio and credentials, an organization schema, and consistent facts across your site and profiles. "Leading provider of…" does nothing; a parseable author entity does.

3

Fix freshness with real updates

Refresh the data, examples, and screenshots, then update the visible date and lastmod. Engines compare content, so a cosmetic date bump without changed content is a wasted (and risky) move.

4

Fix answer-fit by completing the answer

Read the cited competitor's passage next to yours. If theirs answers price, limits, and exceptions while yours answers only price, completeness, not quality, is why they win. Close the gap in one self-contained passage.

05

Verification Checks

How to know it's really done

0/3 verified

The rerank is working for you when near-misses convert. Click a check to mark it verified:

06

Works Together With

The nodes this one leans on

Tools for this from How AI Engines Work

Go deeper from The RAG Pipeline

Always current

These links resolve live: what you get is generated or filtered the moment you click.