Loop Type Doesn't Predict SEO Trajectory: Six Public Companies, 2021 to 2026

Meriin Labs · original research · 14 June 2026 · every number below is computed by published code. See Reproduce this.

The claim: “SEO is a compounding loop, not a channel” is half right. Whether organic traffic compounds or collapses is not set by the loop type. Two companies running the same loop went opposite directions. What would prove it wrong: companies clustering tightly by loop type, all UGC compounding and all content declining.

The popular growth framing says products with a loop (users create content, content attracts users) compound, while content treated as a channel eventually decays. We tested it the boring way: pick six public companies across three loop types, pull five years of monthly organic traffic from Ahrefs, and see whether the loop predicts the curve. It does not. The clearest pair in the set runs the same loop and split by 16x: Reddit grew 16.75x while Stack Overflow, also user-generated content, ended essentially flat at 1.05x and 34% below its own peak.

Hypothesis

Pre-registered before pulling the data. If loop type drove trajectory, companies would cluster by it: UGC and marketplace loops compounding, editorial content declining. Falsifier (and what we found): wide divergence within a loop type.

Dataset

• Companies (6): two UGC loops (reddit.com, stackoverflow.com), two marketplace loops (zillow.com, tripadvisor.com), two editorial/content loops (nerdwallet.com, investopedia.com).
• Metric: monthly organic traffic, Jan 2021 to May 2026 (the final partial month is dropped), from Ahrefs Site Explorer, saved verbatim per company.
• Trajectory: for each, growth multiple (latest 3-month mean vs first 3-month mean), drawdown from the series peak, and CAGR.
• Collected: 2026-06-14. Provenance: Public data only. Public companies, public Ahrefs metrics. No client data.

Method

1. For each company, load the organic-traffic series and drop the final partial month.
2. Compute start (first 3 months), latest (last 3 months), peak, growth multiple, and drawdown.
3. Classify the shape from the numbers: compounding (growth >= 1.3x and within 15% of peak), declining (more than 30% below peak), else flat.
4. Group by loop type and look for divergence.

Loop labels are assigned by business model, not by the data. The shape classification is purely mechanical.

Results

Finding 1: the same loop, opposite outcomes. UGC produced both the biggest winner and a flat-to-declining laggard. Marketplace also split, one compounding and one down 52% off peak. Loop type is not destiny.

Finding 2: only editorial moved together, and it moved down. Both editorial sites declined, losing 60% and 70% of their peak organic traffic. The one place the loop label “worked” is the place where it predicted a loss.

Confidence

• Trajectories (Tier-A). The growth multiples, drawdowns, and shapes are computed from full Ahrefs series. Reddit really did 16.75x; the editorial pair really did lose most of their peak.
• Loop labels (Tier-B). Assigned by business model. A different taxonomy would group the companies differently, but the divergence within any reasonable grouping is the point.
• The “why” (interpretation, not measurement). Our read is that 2024 to 2026 rewarded loops AI amplifies (Reddit became a preferred AI citation source; Zillow’s listing data is proprietary and non-substitutable) and punished loops AI substitutes (Stack Overflow’s Q&A and editorial how-to content are exactly what chatbots now answer directly). That is a hypothesis the data is consistent with, not a proof.

Limitations

• Six companies. A deliberately small, illustrative panel, enough to falsify “loop predicts trajectory,” not to quantify base rates.
• Traffic, not revenue. Organic traffic is a proxy; business outcomes differ (Reddit’s traffic surge coincides with monetization questions of its own).
• One metric, one window. A single five-year organic-traffic series per company; algorithm updates and the AI transition are confounded.
• Loop labels are coarse. Real companies run several loops at once.

How this compares to prior work

Growth-loop thinking is a useful lens, and this is not a refutation of it. It is a correction to the lazy version that says “build a loop and SEO compounds.” On this panel, the loop label explained almost nothing; what tracked with the trajectory was the relationship between the loop and AI answers. The practical move is to ask not “do I have a content loop” but “does an AI answer replace the reason someone visited my loop, or cite it?” Stack Overflow and the editorial sites lost the first test. Reddit and Zillow passed it.

Reproduce this

In data/seo-loops/:

• raw/*.json: each company’s monthly organic-traffic series, saved verbatim from Ahrefs with provenance.
• analyze.py: computes growth, drawdown, CAGR, and the shape per company.
• trajectories.csv, FIGURES-LEDGER.csv: the results and every number in this article.

Run python3 analyze.py. Swap in different companies by adding raw files with a loop_type.

The number

Six public companies, three growth-loop types, five years of organic traffic. Loop type did not predict the trajectory: two UGC sites diverged by 16x (Reddit +16.75x, Stack Overflow flat and 34% off peak), two marketplaces split (Zillow +1.62x, Tripadvisor 52% off peak), and only the two editorial sites moved together, both down 60 to 70% off peak. The loop is not destiny; its relationship to AI answers is.

Changelog & validity

• Valid as of 2026-06-14. Reflects Ahrefs organic-traffic estimates pulled that date.
• A larger panel and a revenue-side view are the planned extensions.
• v1.0, 14 June 2026: first publication under the Meriin Labs team byline. Six companies, figures-ledger written.

Want your own loop pressure-tested against AI substitution? → Book a growth audit

Related: AI engines barely agree on sources · More from Meriin Labs