Which Ahrefs metrics to trust: a Claude Code skill

Reliability is not completeness. In one set of 600,000 keywords only 18% had a CPC, but where it exists it shows roughly what advertisers pay. A sparse column is a sample, not garbage. A fully filled column can still be fiction.

Install it in Claude Code

The file is a ready skill. Its header tells Claude Code what it is and when to load it:

---
name: seo-data-trust
description: How to treat Ahrefs columns and SEO data (KD, DR, Volume, CPC, Traffic value, SERP features, AI Overviews, Search Console). Load before any work with keyword exports or SEO APIs.
---

• For all your projects: save it as ~/.claude/skills/seo-data-trust/SKILL.md.
• For one project: save it as .claude/skills/seo-data-trust/SKILL.md inside the project folder.
• Check that it loads: start a new session and ask Claude which skills it has. seo-data-trust should be on the list. From then on the agent picks it up by itself whenever you give it a keyword export.

Not using Claude Code? Paste the file into your agent’s instructions or a project knowledge file. It’s plain text.

Three levels of trust

Level	What belongs here
Facts	Who is in the live results today. A search feature seen on the snapshot date. Your own Search Console, analytics and click data. Build decisions on these.
Signals	The shape of the 24-month trend. Order of magnitude of volume and CPC. DR near 0 for a site that ranks. Lowest DR in the top ten — only where a snapshot exists. Shortlist with them, with a caveat.
Noise	KD. Traffic value. Organic traffic and Traffic potential as absolutes. DR in the 15–70 range. Don't decide on these.

How to read each Ahrefs metric

Column	How to read it
Volume	12-month average. Order of magnitude only. Seasonality and one-month spikes are hidden — look at the monthly curve first.
SV trend	The most reliable computed metric. Read the shape: growing, fading, one spike, or suspiciously flat.
KD	Links to the top ten only. It fails both ways: 0 with giants in the top, high with a weak top. KD looks too low? Check who ranks. One rough use: sorting thousands of rows.
DR	A real 0 is informative: no links, still ranks. An empty cell is not a 0. Subdomains of blogging platforms inherit the platform's DR. 15–70: don't compare, don't buy links without a manual check.
CPC	Historical snapshot. $1+ vs $0.03 is the signal; the exact figure isn't.
Traffic potential	Relative only. Much higher than volume — a wide cluster. Much lower — zero-click or narrow intent. Empty — no snapshot.
Organic traffic	Compare sites with each other, not absolutes. One junk query can make up a site's whole figure.
SERP features	A fact on the snapshot date. "No AI Overview" is unreliable.
Last update	Mandatory in every export. All results-dependent fields are valid only for this date.
Growth	Noisy on small numbers: 10 → 30 is "+200%".

Nine rules for exports

1. Search results data isn’t captured for all keywords. Empty means “not covered”, not “no feature”. Calculate shares only where a snapshot exists, and call it a lower bound. Dividing by all rows once turned a real 1.77× into a reported 5×.
2. Real difficulty is in the live results. Pull them for the shortlist and save everything.
3. Intent beats keyword metrics. Don’t cut lists mechanically.
4. Conclusions only from data, labelled: confirmed, interpretation, needs checking.
5. New country: export 50–100 known keywords first and measure the coverage.
6. Link growth: look no earlier than 2–4 weeks later; check “live” links by hand.
7. Ahrefs is first a source of keywords in bulk — up to 150,000 rows per export on the Enterprise plan.
8. Search Console shows only organic search, and not all of it. Impressions are not a result.
9. Before trusting a volume figure, look at its 24-month curve. One spike month can show up as a steady “1,200 a month” for a whole year.