Topic Coherence: The Top-Weighted Criterion That Gates Your Entire Score

What this article answers

• What is topic coherence and why is it the top-weighted criterion in AEO Site Rank?
• How does the coherence gate work and when does it cap my overall score?
• How do I improve my blog’s topic coherence quickly?

Key takeaways

• Topic Coherence carries 10% raw weight - tied with Original Data as the single heaviest-weighted criterion in the framework.
• When coherence is measured at 3/10 or lower, a gate activates: your maximum possible AEO Site Rank becomes 35 + (coherence_score * 5). Score of 3 means you’re capped at 50; 4 or higher means no cap.
• The scorer reads the title and first H1 of a fixed sample of up to 12 content pages, building term frequency maps, bigram clusters, and scatter detection to measure thematic focus.
• Words from your domain name and common words like ‘the’ or ‘how’ are filtered out automatically, so only topic terms count.
• Fix this by removing or noindexing off-topic blog posts, then building intentional 5-7 article clusters around your core themes.

The Criterion That Outweighs Everything Else

Topic Coherence carries 10% raw weight inside the Answer Readiness pillar and ties Original Data for the heaviest individual criterion in the entire AEO Rank framework.

At 10% raw weight inside the AEO Site Rank, Topic Coherence is tied with Original Data as the heaviest individual criterion in the framework - and it sits inside the 45% Answer Readiness pillar. It weighs more than Content Depth (7%) and most other criteria, and the coherence gate gives it disproportionate downstream impact across the rest of your score.

Why this much weight? Because AI engines don’t just evaluate individual pages. They evaluate domains. When ChatGPT decides whether to cite your site for a question about customer support software, it doesn’t just read the one relevant article. It scans your blog, your content inventory, your entire topical footprint. And if that footprint looks like a yard sale - AI tools, smoothie recipes, marathon training, office furniture - the engine concludes you’re not an authority on anything.

The scorer works from a fixed sample of up to 12 content pages from your sitemap: articles and posts only, with index, tag, category, listing and service pages left out. The pages are picked in a stable order based on a hash of each URL, so an unchanged site gets the same sample on every run and one newly edited page cannot swing the score. From each page it reads the title and first H1, then builds term frequency maps, extracts bigrams (two-word phrases), identifies clusters, and measures how tightly your content orbits around identifiable themes. Words from your domain name and common words such as “the” or “how” get filtered out, so only topic words count.

A site like Tidio, where every blog post connects to live chat, customer support, or conversational AI, scores high. A site that blogs about whatever the content team felt like writing that week scores low. The math doesn’t care about individual article quality. It measures thematic discipline.

The Coherence Gate: Why a Low Score Caps Everything

A measured coherence score of 3 or lower triggers a hard cap on your AEO Site Rank, with the formula max_score = 35 + coherence × 5 limiting even perfectly executed sites.

Topic Coherence has a unique mechanic no other criterion shares: the coherence gate. When your coherence score is measured at 3 out of 10 or lower, a hard cap activates on your entire AEO Site Rank.

Since October 2026 (calibration v5.1), the gate applies only at 3/10 or lower. Before that, any score below 6 capped the Rank: a 4 held a site to 55 and a 5 held it to 60. A study of 634 sites found no link between that wider cap and AI citations, so scores of 4 and 5 no longer cap. A coherence that could not be measured (fewer than 3 readable pages) never caps.

The formula, applied only when coherence is 3 or lower: max_score = 35 + (coherence_score * 5)

What that means in practice:

• Coherence score 3 = cap at 50. Perfect schema, perfect llms.txt, perfect everything, still capped at 50.
• Coherence score 1 = cap at 40. You’ve effectively disqualified yourself.
• Coherence score 4 or 5 = no cap. The weak score still costs you points through the criterion’s own weight.
• Coherence score 6+ = no cap. Full score potential unlocked.

This gate exists because topical authority is foundational. A scattered blog undermines every other signal. If AI engines can’t figure out what your site is about, they won’t cite it for anything - regardless of how technically perfect your markup is.

Low coherence also limits Content Cannibalization, through a separate rule the October 2026 change left as it was: at 4/10 or lower, the cannibalization score is capped at 6, because scattered content is more likely to produce competing pages that confuse AI about which page to cite. Content Depth is capped at 7 at the same threshold.

Translation: if your AEO Site Rank seems stuck at 50 or below despite strong technical scores, check your coherence first. It might be the gate holding you down.

Only the lowest band caps the AEO Site Rank. The other bands move the Rank through the criterion’s own weight, without a cap.

How the Scorer Measures Coherence

The scorer reads only the title and first H1 of each page in the coherence sample. From those it counts shared words and two-word phrases, measures what share of pages carry the dominant topic, and counts how many separate topic clusters the sample splits into.

The topic coherence algorithm works in four steps:

1. Term Frequency Analysis Every page in the sample contributes its title and first H1; body text is not read. The scorer counts how many pages each word appears on, after dropping words of two letters or fewer, numbers, common words such as “the,” “how” and “your,” and words from your domain name. High-coherence sites show one term appearing across most pages. Low-coherence sites show a flat distribution where no term dominates.

2. Bigram Clustering Single words are ambiguous. “Support” could mean customer support, structural support, or emotional support. Bigrams, two-word phrases like “customer support,” “live chat,” or “ticket resolution,” are far more diagnostic. The scorer collects the bigrams in each title and H1, keeps the ones that appear on at least two pages, and groups bigrams that share a word into one topic cluster.

3. Dominant Topic Share The largest part of the score, up to 7 of the 10 points, comes from the share of sampled pages whose title or first H1 contains the most common topic word or two-word phrase, whichever share is higher. 80% of pages or more earns 7 points, 60% earns 6, 45% earns 5, 30% earns 4, 15% earns 2, and anything lower earns 1. There is no separate check on each page: a page that never mentions the dominant topic simply lowers the share.

4. Scatter Detection The scorer then adjusts for the number of topic clusters from step 2. With a sample of up to 12 pages, 3 clusters or fewer adds 3 points. 4 to 6 clusters adds 1 point, or 2 when at least 70% of pages share the dominant word. 7 to 10 clusters adds nothing, or 1 point for a site that focused. More than 10 clusters takes 2 points away, unless the site is that focused, in which case it adds nothing. The two parts add up to the criterion score, kept between 0 and 10.

The minimum requirement is 3 readable pages. When fewer than 3 articles turn up, the sample is topped up with the site’s other non-listing pages, so a shop or service site without a blog still gets a reading. If fewer than 3 pages can be read even then, the site gets a default mid-range score since there isn’t enough data to measure coherence. That default counts as not measured, so it never triggers the gate.

Fixing a Scattered Blog

Audit existing posts by primary topic, noindex or remove off-topic content like holiday recaps, then build intentional clusters of five to seven interlinked theme articles.

Step 1: Audit your existing content List every blog post. Tag each one with its primary topic. If you can’t describe your blog’s theme in one sentence, you have a coherence problem.

Step 2: Identify and remove off-topic content That company picnic recap, the random holiday greeting, the “10 productivity tips” post that has nothing to do with your business: these are coherence killers. Either delete them, or noindex them and take them out of your sitemap. The scorer samples pages listed in your sitemap, so a post that stays listed can still be picked for the sample.

Step 3: Build intentional topic clusters Pick 2-3 themes directly related to your product or expertise. For each theme, plan a cluster of 5-7 articles that interlink. Example for a customer support platform:

• Cluster 1: Live Chat Implementation (pillar + 5 children)
• Cluster 2: AI in Customer Support (pillar + 6 children)
• Cluster 3: Support Metrics & ROI (pillar + 5 children)

Step 4: Interlink within clusters Every article in a cluster should link to the pillar and to 2-3 sibling articles. The coherence score itself does not read links, since it looks only at titles and first H1s, but interlinking feeds the Internal Linking criterion and helps AI engines see the cluster as one body of work.

Step 5: Establish a content calendar that maintains focus The most common coherence failure is gradual drift. A site starts focused, then someone writes about industry news, then general business advice, then team events. Each post dilutes the signal. Your editorial calendar should enforce topical boundaries.

The fastest path to improvement: audit, remove outliers, cluster what remains. Most sites can jump 3-4 points on coherence in a single focused session of content reorganization.

External Resources

• Topic Modeling for Content Strategy - https://www.semrush.com/blog/topic-clusters/
• Content Clustering Best Practices - https://ahrefs.com/blog/content-hub/
• Google’s Helpful Content Update and Topical Authority - https://developers.google.com/search/docs/fundamentals/creating-helpful-content

Related topics

• content-depth
• original-data
• internal-linking