Perplexity Favors Fresh Content: How to Measure the Freshness Signal

Time series chart with rising data trend

The Claim

Of the four major AI engines we track at GEOlytic, Perplexity is the only one whose answers visibly degrade if you stop publishing. ChatGPT's training corpus carries it through publishing droughts. Claude relies heavily on canonical primary sources. Gemini leans on traditional Google signals. Perplexity, by contrast, runs live retrieval against the open web on every query — and our analysis (and the published industry research below) suggests it weights very recent content disproportionately.

The number floating around in industry research: content published in the last 30 days carries roughly a 3.2× boost in Perplexity citation likelihood vs. content older than 90 days, holding domain authority and topical match constant.

If that's right, it changes how you plan a content calendar. It means your "evergreen guide from 2024" is a depreciating asset on Perplexity even when it's still ranking on Google. And it means a steady cadence of fresh, dated, citable posts is a real moat — one that traditional SEO tools don't measure.

This post is the methodology for testing that claim on your brand over a 30-day window, plus what the published data says before you start.

TL;DR: Set up two cohorts of equivalent prompts. Track Perplexity citation rate before and after a fresh content drop. Compare the lift against a control group with no new content. Industry priors say expect a 2-3× lift on the prompts where you publish; we'll publish our own results in the next post.

What the Industry Research Actually Says

Before running an experiment, anchor on what's been published. The strongest sources for Perplexity-specific freshness signal:

| Source | Finding | |---|---| | Otterly Citation Economy Report 2026 | Perplexity citations show 53.4% monthly turnover — meaning more than half the brands cited this month are different from last month. The churn is driven primarily by content recency. | | Profound's GEO Guide 2025 | Perplexity's retrieval pipeline preferentially pulls from URLs with <lastmod> dates within 30-90 days. Older content needs significantly higher domain authority to surface. | | Aggarwal et al., KDD 2024 | Original-data and statistics-heavy content outperforms generic content by 41% in citation likelihood across answer engines. Freshness amplifies this — new statistics outperform old statistics by another ~28%. | | Maximus Labs — Perplexity SEO Guide | Perplexity's algorithm gives roughly 3.2× higher citation probability to content under 30 days old vs. content over 90 days, on equivalent queries. |

Three things stand out:

The freshness signal is multiplicative, not additive. A fresh post on a high-authority domain compounds. A fresh post on a brand-new domain still gets a lift, but from a much smaller base.
<lastmod> matters as much as datePublished. Updating an existing post (real updates, not date-touching) appears to recover most of the freshness lift.
The freshness window is ~30 days for a peak, ~90 days for a slow decay. Past 90 days, Perplexity treats content as effectively "evergreen" and falls back to authority signals.

That last point is what makes the 30-day experiment design so clean. The decay curve gives you a clear before/after window to measure against.

The 30-Day Experiment Design

Step 1: Pick Your Prompt Cohorts

You need two cohorts of 10-15 prompts each, matched on intent and category but probing slightly different facets of your business.

Cohort A (publishing target): Prompts where you'll release new content during the experiment. Example for a project management SaaS: "What are the best project management tools for remote teams?", "How do I track team capacity?", "Best alternatives to Asana?"
Cohort B (control): Equivalent prompts in your category where you'll NOT publish anything new during the 30-day window. Same broad intent, different question stems.

The matching is what gives the experiment its statistical bite. Without a control, you can't tell whether observed lift came from your freshness drop or from some unrelated change in Perplexity's index.

Step 2: Measure Baseline (Days -7 to 0)

Run each prompt across both cohorts daily for the week before publishing. Record:

Mention rate: % of responses that name your brand
Citation rate: % of responses where Perplexity cites your domain as a source (the [1]-style footnotes)
Position: where you appear (first sentence, mid-paragraph, footnote-only)

A daily run gives you 7 data points per prompt per cohort. That's enough to establish a noise floor for each prompt.

If you're running this on GEOlytic, scheduled daily AIVS runs handle this automatically. Otherwise: a simple script that hits Perplexity's API + parses the citations field is ~50 lines.

Step 3: Publish Fresh Content (Day 0)

On Day 0, publish 3-5 high-quality posts targeting Cohort A's prompts. Real content, not date-touching. Each post should:

Have a clear publish date in the URL or metadata
Include datePublished and dateModified in Article schema (or NewsArticle if it's commentary on a recent industry event)
Cite specific data, sources, and statistics — Perplexity's retriever appears to preferentially surface posts with concrete factual density (per the Aggarwal et al. finding above)
Be genuinely useful, not stuffed with brand mentions — the goal is for Perplexity to cite your domain as the source of an answer, not for you to spam your name

Don't touch Cohort B. Don't even update old posts in Cohort B's territory. The control has to stay still.

Step 4: Measure Lift (Days 1 to 30)

Continue daily probing. Look for:

| Metric | Expected Cohort A lift | Expected Cohort B lift | |---|---|---| | Mention rate | +50-200% by Day 7-14, peak Day 14-21, slow decay after | ±10% noise — ideally flat | | Citation rate | +200-400% by Day 7-14 (citation rate compounds harder than mention rate) | ±10% noise | | Position | Move from footnote-only to in-paragraph or first-sentence on a meaningful share of responses | No movement |

If Cohort A shows the predicted lift and Cohort B is flat, you've measured your domain's freshness coefficient. If both cohorts move together, something else changed (Perplexity index update, news event, competitor publish) and you'll need to re-run.

Step 5: Decay Tracking (Days 30 to 90)

After Day 30, don't publish anything new on Cohort A's territory. Keep probing. The shape of the decay curve is your real takeaway:

A steep decay (back to baseline by Day 60) means freshness is your primary lever — you need a permanent publishing cadence
A shallow decay (still elevated at Day 90) means the new posts are accruing authority signals beyond freshness, and your domain is starting to compound
A re-spike around Day 45-60 (without new posts) usually means a competitor's content got de-indexed and you backfilled

What This Means for Your Content Calendar

The implications for a small team are uncomfortable: continuous publishing matters more than long-form effort, at least for Perplexity-specific visibility.

A team publishing 3 quick 800-word posts a week will outperform a team publishing one heroic 5,000-word guide a month, even if the guide is objectively better content. The delta isn't because the short posts are better — it's because Perplexity's freshness boost makes them more retrievable during their 30-day peak, and the cadence keeps something always in-window.

The right model is probably hybrid:

Evergreen guides (like the post you're reading) — written for long-tail SEO and ChatGPT/Claude citation, where authority and structure dominate
News-style posts with NewsArticle schema — short, dated commentary on real events (model launches, ranking shifts, competitor moves). One per week minimum to keep the freshness window populated.
Quarterly evergreen refreshes — bumping dateModified on existing posts after meaningful updates, which appears to recover most of the freshness lift without writing a new post

The mistake is to do only the first kind. The second kind is what the 30-day experiment is designed to validate.

What We're Doing at GEOlytic

We're running this exact experiment on geolytic.ai over the next 30 days. Two cohorts of 12 prompts each (one publishing, one control). New content drops on Days 0, 7, 14, and 21. Daily probing across all four engines, not just Perplexity, to see whether the freshness signal is uniquely Perplexity-shaped or whether ChatGPT/Claude/Gemini also reward recency.

The results — including the prompt list, raw probe data, and the actual lift curves — will be the next post in this series. If the published priors hold, expect a Perplexity citation lift of 2-3× on Cohort A by Day 14, decaying to ~1.5× by Day 60. ChatGPT and Claude are expected to be much flatter; Gemini is the wild card.

If you want to run the experiment on your own brand alongside us, GEOlytic's scheduled AIVS runs handle the daily probing and citation tracking automatically — which is the part that's tedious to script manually. Sign up for the beta and we'll share our prompt-cohort template.