Evidence-based SEO
How to run SEO as measurement rather than opinion: the competitive set, the practices with their evidence tiers, the technical floor, and the tactics the product refuses to automate.
38 lessons in 6 modules · for marketers and consultants who want advice they can defend · each lesson shows its evidence tier or its basis
Academy Evidence-based SEO
- 01Operate on evidence5
- 02The competitive set4
- 03Content that correlates with page 18
- 04The technical floor7
- 05Links and brand, in proportion4
- 06What we never do: 10 excluded tactics10
Open a lesson to read what to do, why, and how the product measures it.
Module 1 of 6
Operate on evidence
The loop: Search Console as feedback, scheduled crawls, controlled tests, a short action list.
What to do: Act on queries a page ranks for but does not cover, positions 4–20 with high impressions, and CTR below the expected curve.
Why (confirmed): First-party data from Google about your own pages.Google states it in documentation or official guidance.
How AwesomeSEO measures it: Opportunity queue from GSC, with your own fitted CTR curve.
Search performanceWhat to do: Run the technical crawl weekly and push new issues into the actions register with an owner.
Why (practitioner): Problems are cheapest when caught the week they appear.A working heuristic from experienced SEOs. We test it on your own data.
ActionsWhat to do: Change a group of similar pages, keep a matched group unchanged, and test the difference for significance before rolling out.
Why (practitioner): Separates the effect of the change from seasonality and updates.A working heuristic from experienced SEOs. We test it on your own data.
Evidence libraryWhat to do: Rank actions by impact and effort, assign an owner, and keep the list short enough to finish.
Why (practitioner): Recommendations that are never implemented have zero value; most SEO programmes fail on execution.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Actions register with owner, status, impact, effort and completion rate.
ActionsWhat to do: Label how AI helped, have a named person review every page, and check AI drafts for accuracy and originality.
Why (confirmed): Google accepts AI-assisted content that helps people; scaled low-value content is a spam policy violation.Google states it in documentation or official guidance.
How AwesomeSEO measures it: Approval log on every agent draft.
AgentsModule 2 of 6
The competitive set
Why the top 20 is the unit of analysis, and what to do at each ranking stage.
What to do: For every money keyword, keep a standing study of the top 20: who ranks, what they cover, what they get wrong, what gain is left.
Why (practitioner): Below the top 20 the order is noisy. The set changes slowly, so a long-running watch shows who is gaining between core updates.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: SERP Lab snapshots, scheduled per keyword.
SERP LabWhat to do: Compare your page to the top 20 metric by metric and close the gaps where you sit outside the podium. Do not chase a single metric to extremes.
Why (practitioner): Advantages compound across many signals. An extreme value on one metric looks unnatural and rarely moves rankings alone.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Podium score: share of metrics where the page ranks in the top 3 of the SERP, outliers flagged.
SERP LabWhat to do: Top 100: on-page relevance. Top 20: links and information gain. Top 10 and top 5: user signals, CTR, brand demand and history.
Why (practitioner): A working model from the SEO manifesto. Click signals at the top are supported by NavBoost testimony in US v. Google (2023).A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Lever recommender per keyword from position, CTR gap, freshness and link gap.
Search performanceWhat to do: Outperform the closest comparable competitor (same page type, same intent) on quality, publishing rate, speed and user signals for months.
Why (speculative): Observed pattern: sites that perform consistently between updates are tested against incumbents at the update and can swap places.A hypothesis. Shown with that label, never as advice.
How AwesomeSEO measures it: Momentum state per SERP (positive, neutral, negative) and closest-rival tracker.
SERP LabModule 3 of 6
Content that correlates with page 1
The measured practices: entities, variants, structure, answers, real freshness.
What to do: Name the real things (products, places, people, models, standards) and describe them through their attributes.
Why (measured): Number of Distinct Entities Used is the most durable strong factor: in the top 200 for 81 of 93 months, average r −0.27.Correlates with position in the TopSEOFactors data. Correlation, not causation.
How AwesomeSEO measures it: Entity count and entity-in-title/H1/H2 against page-1 averages.
- Number of Distinct Entities Used #45 · r −0.18
- Entities in Title Tag #73 · r −0.17
- Entities in H2 Tags #44 · r −0.18
What to do: Cover related terms, jargon and attribute words naturally. Do not stuff them into the title.
Why (measured): Unique LSI words used is #5 (r −0.24), while LSI words in the title fell 110 places and clean keyword density fell 98.Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Number of Unique LSI Words Used #5 · r −0.24
- LSI Words in Title Tag #150 · r −0.14
- Clean Keyword Density in the HTML Tag #134 · r −0.14
What to do: Collect the words Google highlights in top-20 snippets (singular, plural, definite forms, synonyms) and use them across headings and body.
Why (measured): Google bolds what it treats as the same query. Unique variations used correlates (r −0.19).Correlates with position in the TopSEOFactors data. Correlation, not causation.
How AwesomeSEO measures it: Variant list from SERP highlights per keyword.
- Variations in H1-H6 Tags #28 · r −0.19
- Variations in H1, H2, and H3 Tags #26 · r −0.19
What to do: One H1, about 7 H2s and 9 H3s on a money page, one H2 per attribute with data.
Why (measured): Page-1 averages in the latest month: H1 1, H2 7, H3 9, heading tags 20 (all r ≈ −0.17 to −0.20, stable for years).Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Number of H1 Tags #22 · r −0.20
- Number of H2 Tags #23 · r −0.20
What to do: Open each section with a direct answer of about 40 words, exact where consensus allows, with the answer in bold. Use real questions as H3s.
Why (measured): Questions in headings correlate (QS004, r −0.16, 82 of 93 months) and direct answers are what featured snippets and AI answers lift.Correlates with position in the TopSEOFactors data. Correlation, not causation.
How AwesomeSEO measures it: Answer-first check per heading in the brief and the audit.
- Number of Questions In Headings #76 · r −0.16
What to do: Add a calculator, finder, comparison table, size guide or quote form above the fold.
Why (measured): Has Form climbed 78 places year on year (r −0.17) and Number of Images is #12 (r −0.21): page-1 results show what they do.Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Has Form #65 · r −0.17
- Number of Images #12 · r −0.21
What to do: Replace "many" and "often" with figures, stay inside the expert consensus, and cite the source.
Why (practitioner): Precise, sourced claims are easier to verify and to quote, which is what both readers and AI answers reward.A working heuristic from experienced SEOs. We test it on your own data.
ContentWhat to do: Revise at least 10–15% of a page when you update it. Annual facts get updated every year, with year (and month) in the title where it helps.
Why (documented): The leak shows several date signals (byline, syntactic, semantic). Changing only the date is not an update.The signal exists per the 2024 API leak, trial testimony or a Google patent. Weight unknown.
How AwesomeSEO measures it: Freshness queue by tier and revision share between versions.
ContentModule 4 of 6
The technical floor
Confirmed by Google or measured in the data: what every template must pass.
What to do: LCP ≤ 2.5 s, INP ≤ 200 ms, CLS ≤ 0.1, measured on real users.
Why (confirmed): Part of Google’s page-experience guidance. INP replaced FID in March 2024.Google states it in documentation or official guidance.
Site healthWhat to do: Server-render or pre-render pages that should rank; check that text, links and schema are in the HTML response.
Why (confirmed): Google renders JavaScript, but later and at a cost; many AI crawlers do not render it at all.Google states it in documentation or official guidance.
Site healthWhat to do: Aim for a server response under 200 ms (100 ms on money pages), clean HTML and few includes.
Why (measured): Load Time Milliseconds is #6 (r −0.23) and gained the most correlation since 2019 (−0.14 → −0.22). Slow servers also cut crawl rate.Correlates with position in the TopSEOFactors data. Correlation, not causation.
How AwesomeSEO measures it: TTFB per template from the crawler and field data.
- Load Time Milliseconds #6 · r −0.23
- Number of Includes #9 · r −0.21
What to do: Organization and WebSite sitewide; Product, Article, BreadcrumbList, ItemList and LocalBusiness where they apply; SoftwareApplication for apps.
Why (measured): Has JSON-LD is #11 (r −0.21); the structured-data pillar grew from 3% to 13% of the top 200 since 2019.Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Has JSON-LD Schema Markup #11 · r −0.21
- Uses JSON-LD name #71 · r −0.17
- Uses JSON-LD applicationCategory avg r −0.12
What to do: Every structured-data claim must be visible on the page.
Why (confirmed): Google’s structured-data guidelines; mismatches can lose rich results or trigger a manual action.Google states it in documentation or official guidance.
How AwesomeSEO measures it: Schema parity check in the crawler: share of JSON-LD values found in visible text.
Site healthWhat to do: One indexable URL per piece of content, 301 for moves, 404/410 for removals, no chains.
Why (confirmed): Google documentation on canonicalisation and redirects.Google states it in documentation or official guidance.
Site healthWhat to do: Split sitemaps by type, keep lastmod truthful.
Why (confirmed): Google uses lastmod when it is consistently accurate and ignores priority and changefreq.Google states it in documentation or official guidance.
Site healthModule 5 of 6
Links and brand, in proportion
Referring domains, a natural mix, and branded search that keeps pace.
What to do: Publish original data and indexes, supplier and partner pages, local press and trade-body memberships.
Why (measured): Number of Referring Domains is the #1 factor in September 2026 (r −0.28); authority holds 4 of the top 10.Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Number of Referring Domains #1 · r −0.28
- Number of Do Follow Backlinks #2 · r −0.27
- Number of Backlinks #3 · r −0.27
What to do: Branded and descriptive anchors, followed and nofollow links, image links, in proportions normal for the niche.
Why (measured): Number of nofollow backlinks is #4 (r −0.25): natural profiles contain them.Correlates with position in the TopSEOFactors data. Correlation, not causation.
How AwesomeSEO measures it: Anchor and link-type distribution vs the niche median.
- Number of No Follow Backlinks #4 · r −0.25
What to do: Social profiles, Google Business Profile, Bing Places, trade directories and Wikidata, with identical name, address, phone and description.
Why (measured): Number of Social Accounts climbed 96 places in a year; Facebook and Instagram page counts correlate.Correlates with position in the TopSEOFactors data. Correlation, not causation.
- Number of Social Accounts #79 · r −0.16
- Number of Facebook Pages #15 · r −0.20
- Number of Instagram Pages #41 · r −0.18
What to do: Build searches for the brand and for brand + keyword through PR, social and product. Watch that link growth does not outrun brand demand.
Why (documented): NavBoost click data and site-level signals reward sites people seek out. A profile with many links and no brand searches looks bought.The signal exists per the 2024 API leak, trial testimony or a Google patent. Weight unknown.
How AwesomeSEO measures it: Link-to-brand balance: referring domains per 100 branded searches vs the niche.
Brand & authorityModule 6 of 6
What we never do: 10 excluded tactics
Each with the reason. The product refuses to automate any of them, whatever a competitor does.
What to do: Never. The product refuses to automate it.
Why (excluded): Why it is excluded: Fake clicks violate Google’s spam policies and click systems filter them.Breaks spam policies or our ethics. Never automated.
What to do: Mark paid placements rel="sponsored".
Why (excluded): Why it is excluded: Link spam under Google’s spam policies.Breaks spam policies or our ethics. Never automated.
What to do: Acquire domains only as real businesses or content you continue.
Why (excluded): Why it is excluded: Expired domain abuse became a spam policy in March 2024.Breaks spam policies or our ethics. Never automated.
What to do: Use hreflang only for real language versions.
Why (excluded): Why it is excluded: Manipulative use of annotations; no evidence it works.Breaks spam policies or our ethics. Never automated.
What to do: Seed discussion with real people only.
Why (excluded): Why it is excluded: Deceptive and against platform rules; worthless as a signal.Breaks spam policies or our ethics. Never automated.
What to do: Outreach agents draft from real, named senders and need approval to send.
Why (excluded): Why it is excluded: Deceptive, and illegal in some markets.Breaks spam policies or our ethics. Never automated.
What to do: One page per real service area.
Why (excluded): Why it is excluded: Doorway abuse under Google’s spam policies.Breaks spam policies or our ethics. Never automated.
What to do: Update at least 10–15% of the page or leave the date.
Why (excluded): Why it is excluded: Misleads readers; Google advises against it.Breaks spam policies or our ethics. Never automated.
What to do: Collect reviews from real customers only.
Why (excluded): Why it is excluded: Breaks Google review policies and consumer law.Breaks spam policies or our ethics. Never automated.
Next track
Practise on the demo workspace: Evidence library.
Example data for a fictional Swedish brand, every number with its source, interval and n.