Methodology
How the Semantic Map works
A plain-English explanation of what this tool does, why it matters for SEO and AI search, and what you should do with the results. Written for anyone — AMs, content strategists, clients, prospects.
The core concept
Think of Google as a librarian trying to figure out exactly what your business is an expert in.
If you own a personal injury law firm, you want Google to see you as the ultimate authority on car accidents, slip and falls, and local courtrooms.
However, over the years, websites tend to accumulate clutter:
- An old blog post about "Top 10 Summer Driving Tips" (drifting into general travel).
- A page about a charity event you sponsored five years ago.
- An outdated page answering a general legal question that has nothing to do with personal injury.
Individually, these pages seem harmless. But collectively, they create topical noise. Over time, Google starts seeing your site as a generic mix of random topics rather than a laser-focused expert in your specific niche.
What is a "semantic map"?
A semantic map is a visual tool that plots every single page on your website as a dot to show how closely related each page is to your main business focus.
Imagine a target with a bullseye:
Your most important money pages — homepage, top service pages, primary location pages. These are what you most want Google associating with your brand.
Highly relevant supporting content — answering questions your customers actually have, covering on-topic sub-areas of your service.
Broadly related topics that aren't your primary service but reinforce the same industry. Often FAQ pages, supporting resources, related practice areas.
Pages that have drifted away from what your business sells. The longer a site has been publishing, the more of these tend to accumulate.
If your map shows too many dots drifting into the outer rings, your website's authority is being diluted.
Why this matters for SEO & AI search
Google doesn't just rank individual pages anymore — it ranks entities and overall site authority. If 30% of your website consists of outdated, off-topic blog posts, Google gets confused about what you actually sell. To rank higher for your core services, your entire website needs to send one unified, crystal-clear message.
The same principle drives AI search. ChatGPT, Perplexity, and Google's AI Overviews all use retrieval-augmented generation — they cite sites whose content most cohesively clusters around the topic being asked about. A site that looks "about" twelve things won't get cited as the authority on any one of them.
A focused site wins twice: better rankings on traditional search, and a higher chance of being the source AI assistants quote when someone asks for an expert.
The action plan
Seeing your site this way lets you make smart cleanup decisions:
Keep & link
Identify your strong Core and Focus pages and make sure they link internally to each other. The tighter the internal-link cluster around your services, the stronger the signal.
Update
Find pages that are drifting slightly (Expansion zone, or Focus pages in a Mixed-diagnosis section) and rewrite them to tie directly back to your main services.
Prune (delete or redirect)
Find the Peripheral pages that no longer serve your business goals. Either delete them entirely (if they get no traffic) or redirect them to a more relevant page. This instantly sharpens Google's understanding of your brand.
Beyond the orbit chart — what else the tool does
The orbit chart is the foundation, but it's only the first layer. Once we've classified every page into Core / Focus / Expansion / Peripheral, several other reports sit on top of that classification — each one stratified by zone, because a backlink to a Core page matters more than a backlink to a Peripheral one, a cannibalization conflict between two Core pages is a five-alarm fire while one between two Peripheral blog posts barely registers, and so on.
AI Search Readiness
What it does: A second score (alongside Topical Health) that estimates how likely ChatGPT, Perplexity, and Google AI Overviews are to cite this site as an authoritative source. Broken into four sub-scores so you can see exactly where the gap is.
- Cohesion — how tightly the content clusters around the core topic
- Depth — median word count across on-topic pages (AI engines favor substantial answers)
- Schema — % of pages with structured data markup (Article, Organization, FAQ, etc.)
- Breadth — number of distinct URL sections covering on-topic content
Why useful: Traditional SEO and Generative Engine Optimization (GEO) aren't the same. A site can rank in Google's blue links but be invisible to AI Overviews. This score tells you which layer needs attention.
Per-page recommendations + Keep/Kill workflow
What it does: For every drifted page, Claude (Anthropic's LLM) reads the actual page content in the context of the site's intended core topic and returns a specific recommended action — Keep, Rewrite, Consolidate, Redirect, Delete, or Noindex — with reasoning and a verbatim phrase from the page proving the AI actually read it (not hallucinated).
Keep/Kill workflow: The AM toggles each page Keep or Kill from the list view. Once you've decided enough pages, hit Download decisions on the client folder to generate a print-ready PDF — grouped by Kill (prune/redirect queue) and Keep (confirmed on-brand), with the AI's reasoning inline. Hand it to the content team.
Why useful: Other SEO tools tell you a page is “low quality.” This one tells you why, in language a content lead can actually act on, with the cited phrase as evidence.
Competitor benchmarking (Compare view)
What it does: Add 2–5 competitor domains under any client folder; the tool runs the same semantic map analysis on each. The Compare view shows them side-by-side with:
- Headline scoreboard (zone counts, AI Readiness grade per site)
- AIO sub-score gap analysis — where you're behind and exactly what to do about it
- Content gap analysis — URL sections competitors have that you don't, ranked by competitive signal
- Strategic moves table — prioritized agency-deliverable list with effort sizing (S/M/L)
- Section coverage matrix — every URL section across every site, with diagnosis badges
Why useful: Pitches, proposals, QBRs. Translates the abstract idea of “competitive positioning” into a one-page deliverable the client can read.
Off-page backlink intelligence
What it does: Pulls Ahrefs backlink data and stratifies it by semantic zone. Two views:
- Link health by zone: a bar chart showing how referring domains are distributed across Core / Focus / Expansion / Peripheral. Healthy sites concentrate authority on Core + Focus.
- Underlinked Core/Focus queue: your most anchor-valuable pages sorted by referring-domain count ascending. The off-page team's weekly priority queue — where outreach effort compounds the most.
Why useful: Ahrefs treats every page on a site as equal — a backlink to an obscure 2019 blog post counts the same as one to your main service page. This view shows you where links are actually compounding for the business.
Content cannibalization analysis
What it does: Detects when two or more of your own pages target the same keyword or cover near-identical content — splitting authority and confusing Google about which to rank. Two layers:
- Keyword cannibalization (Ahrefs): keywords where 2+ of your pages are already ranking in the top 30. The confirmed-conflict tier.
- Semantic near-duplicates (predictive, free): page pairs with cosine similarity ≥ 0.92 — pages that look like near-duplicates to a language model, regardless of whether Google has noticed yet. Catches conflicts before they split rankings.
Pick winner: One click on the URL that should keep ranking → that URL gets tagged Keep, the others Kill, all flowing into the Keep/Kill PDF.
Why useful: Cannibalization is one of the most common silent killers of rankings on sites with 3+ years of content. The semantic-similarity layer is the differentiator — no other tool does that, because no other tool has the embeddings.
Run history (drift over time)
What it does: Every client folder keeps a complete history of every analysis run. Quarter to quarter, you can see whether peripheral count is shrinking (you're cleaning up), the score is climbing (the team's work is moving the needle), or drift is growing (something's slipping).
Why useful: Retainer justification. The orbit chart from January next to the orbit chart from April is the most honest QBR slide an agency can produce.
Where everything lives in the UI
Quick orientation so you know where to click for each feature.
| Feature | Where to find it |
|---|---|
| Orbit chart | Run page → Chart tab |
| Page list with per-page recommendations | Run page → List tab |
| Cannibalization (keyword + semantic) | Run page → Cannibalization tab |
| Topical Health Score + AI Readiness | Run page → top of page (score cards) |
| Strategic roadmap / insights | Run page → below the chart |
| Keep/Kill PDF download | Run page → Download decisions button |
| Branded client report | Run page → Client report button |
| CSV export of every page | Run page → Export all button |
| Seed URLs (pinned centroid) | Client folder → seed URLs editor |
| Competitor benchmarking | Client folder → Compare side-by-side |
| Off-page backlink intelligence | Client folder → Off-page panel |
| Run history (drift over time) | Client folder → Run history section |
| All clients overview | Clients tab in the top nav |
Behind the scenes
If you want to know exactly how the tool produces those dots and rings, here's the pipeline.
- 1
Crawl
Starting from the site's sitemap.xml (or the homepage if no sitemap exists), we walk through every indexable page, respecting robots.txt and rate-limiting politely. Capped at 1,000 pages per run.
- 2
Extract main content
For each page, we strip out navigation, footer, ads, and other site chrome. What's left is the actual content the page is about — the same content a careful reader (or Google's ranking algorithm, or ChatGPT's retrieval system) would care about.
- 3
Embed each page
We send the extracted content through OpenAI's text-embedding-3-small model. Out comes a list of 1,536 numbers — a mathematical fingerprint that captures the page's meaning. Two pages about similar topics produce similar fingerprints. Two pages about wildly different topics produce very different fingerprints.
- 4
Compute the site center
The site center is the average of every page's fingerprint — mathematically, the middle of what the site is about across all its content. (For multi-business-line sites, you can pin the center to specific seed pages instead — see the Pinned Centroid mode.)
- 5
Measure distance
For each page, we compute how far its fingerprint is from the site center. Pages close to the center reinforce the brand's main topic; pages far away have drifted off-topic.
- 6
Bucket and plot
We sort each page into one of the four zones (Core / Focus / Expansion / Peripheral) based on its distance, then plot every page on an orbit chart. The radius from the center is the actual distance; the angle around the center groups semantically-similar pages together, so you often see clear clusters of related content.
The Topical Health Score
The 0–100 score (and letter grade A–F) at the top of every run is computed deterministically from the bucket distribution. Peripheral % is the dominant penalty; Core % gives the biggest bonus.
score = 100 − (peripheral% × 1.0 + expansion% × 0.3 − core% × 0.5 − focus% × 0.2)
Clamped to 0–100. Letter grade: A ≥ 85, B ≥ 70, C ≥ 55, D ≥ 40, F < 40.
Because the formula is fixed, the score is comparable across clients. A 78 means the same thing for every site — meaningful portfolio comparisons, fair quarter-over-quarter tracking.
Important: "peripheral" doesn't always mean "off-topic"
The site center is the average of every page on the site. If a site has multiple business lines and one has more pages than others, the smaller business lines will look peripheral simply because they're under-represented in the average — not because they're irrelevant to the business.
Example: a multi-practice law firm with 800 personal-injury pages and 200 family-law pages will see the family-law pages flagged as peripheral, even though they're a legitimate part of the business. The center has been pulled toward personal injury by sheer volume.
The fix is Pinned Centroid mode. Designate a handful of seed URLs (homepage + top service pages, one per business line) and the tool will measure distance against those instead of the floating average. For multi-practice firms, this is the more honest read.
When to run an analysis
- New client onboarding: baseline audit in week 1. Shows the client what they're working with and where the opportunity is.
- Quarterly health check: re-run to prove drift is shrinking — or catch it growing before it costs rankings.
- Pre-publishing strategy: before greenlighting a new content silo, see whether existing peripheral pages need cleanup first.
- Pitch decks and proposals: the orbit chart visualizes a problem most agencies can only describe.
Honest limitations
- The tool measures meaning, not quality. A poorly written but on-topic page lands near the center; a beautifully written but off-topic page lands on the edge. We're measuring the footprint, not the craft.
- It doesn't predict rankings directly. A high score correlates with strong topical authority, but rankings also depend on backlinks, technical SEO, content depth, and other factors.
- Cap of 1,000 pages. Larger sites are sampled by sitemap order. A future sampling option will let you pick top-traffic pages or filter by URL pattern.