Loading data
Preparing the interface and the latest data

SEO text relevance analysis

Analyze relevance against the top-results median and examine your page content in depth.
Compare your page's SEO metrics with the leading competitors in the top 10.
Entities, search intents, semantic models, unigrams and more.
0 characters

For English content, singular/plural and other word forms are counted separately. Russian lemmatization is unchanged. English content supports the main relevance analysis and document similarity, but automatic page-profile section inference is not yet available. Advanced semantic block classification requires Russian content and Russian report language.

Saved report text and source excerpts keep their original language.

Run and sample

Competitor source The first analysis gets competitors from search results. Turn off this setting before refreshing to get a new sample. Manually entered pages always remain fixed.

Additional features

Text cleaning

Blocks and exclusions

0 characters
0 characters
0 characters
0 characters
0 characters
0 characters
0 characters

❓ Frequently asked questions

Key points about the service and how the analysis works.
What is hybrid relevance analysis and how does it work?

Hybrid relevance analysis compares the documents you provide using a hybrid method. Compare your page with top-ranking competitors from Yandex or Google, or manually enter any page's HTML. The analyzer evaluates how closely the page content matches the consistent semantic model of the niche.

The service compares your page with competitors to identify the terms, meanings and structures that form the market standard.

🧠 How to check website or text relevance

The system works through several stages:

  • cleans HTML of scripts, styles, navigation and other noise
  • extracts text and splits it into terms and phrases
  • normalizes Russian word forms through lemmatization; English forms are counted separately
  • compares term frequencies with competitors using TF-IDF
  • builds a reference model for your niche

📊 What is analyzed

  • Topic coverage — terms used in the niche
  • Occurrences — how frequently key terms are used
  • Structure — how content is distributed across blocks
  • Anchor list — terms used in links

📦 Example

Suppose the topic is SEO services.

Competitors frequently mention:

  • “technical audit”
  • “keyword research”
  • “on-page optimization”
  • “link building”

If your page does not contain these terms, they are shown as missing.

If they appear but are used less frequently, they enter the recommendations for occurrences.

🎯 What you get

  • a list of missing terms
  • recommendations for improving the text
  • an understanding of a suitable page structure
  • comparison with competitors

⚠ Keep in mind

The service analyzes more than words: it examines semantic units and market patterns.

It is a tool for building an appropriate page model.


Put simply:

  • competitors → form the niche reference
  • your page → is compared with that reference
  • result → what to add or improve
How do I check page relevance and use the results?

What the check shows

The check shows how your page's content and structure compare with pages the search engine considers relevant to the selected query and region. It evaluates topic coverage, term occurrences, HTML zones, questions, entities and page types in search results.

How to run an analysis

Enter one target query, choose a search engine and region, then add the page URL. Compare pages with the same purpose: services with services, categories with categories, articles with articles. For mixed search results, consider the COMM, INFO and MIXED labels when interpreting the results.

How to use the results

First review the documents included in the top-results reference, then examine consistently missing terms, occurrence recommendations and HTML-zone contributions. Add useful sections that develop the relevant meaning and answer the user's question, rather than isolated words to increase a score.

After updating the page, run the analysis again and compare coverage, occurrences and overall relevance. Make changes in stages so you can identify which edits improved the result.

The score does not guarantee a ranking: site quality, links, commercial factors and other signals also affect search results.

How is the semantic model of top results built?

The top-results semantic model is a page model derived from competitors. It examines how the market covers a topic: entities, questions, important sections and the similarity between your page and the overall search intent.

The model turns recommendations into practical actions: add a block, answer a question, rewrite a section, separate irrelevant content or sharpen the page's topical focus.

Competitors establish a market reference, your page is compared against it, and the service identifies semantic gaps.

How it is calculated

  • HTML is collected from your page and competitors;
  • pages are cleaned of technical noise, scripts, styles and unnecessary blocks;
  • entities, questions, headings, structure and page profiles are extracted;
  • embedding vectors are calculated for documents and questions;
  • your page is compared with the average competitor model and individual competitors;
  • the gaps are used to assemble explanations and an edit plan.
How is page relevance calculated?

The algorithm builds a comparative model from several competitors.

It considers:

  • term frequency,
  • importance within the topic,
  • consistency across different market participants,
  • and how thoroughly the concepts are covered.

Overall relevance balances topic coverage with the depth of treatment of key terms.

The report also separates market coverage, consistent-core coverage, similarity of signal distributions and rare elements requiring manual review. This helps identify specific weak areas rather than relying on one overall percentage.

What do INFO, COMM and MIXED labels mean in the document table?

The service estimates each page's type from its content and structure. Labels help explain the kinds of pages the search engine returns for the query:

  • INFO — an informational page: article, guide, review or reference material.
  • COMM — a commercial service, category or offer page.
  • MIXED — a page combining commercial and informational content.

Informational pages are not automatically removed. They reflect real search results and show related information the search engine may expect. You do not need to turn a commercial page into an article: use these pages as sources of questions, terms and explanations, while comparing primary metrics with pages of your own type.

Classification is a diagnostic estimate by the service, not an official search-engine label.

How close is your page to the intent of top results?

Document cosine similarity compares your page's embedding vector with competitor vectors and the average top-results model. It reflects the overall topic, rather than individual term frequency.

A high percentage indicates that the page broadly addresses the same intent as competitors. A low percentage may indicate an adjacent topic, several services mixed together or a structure unlike the top results.

Reading the values

  • 82% or higher — the overall intent is aligned; address individual gaps.
  • 70–81% — the topic is similar, but parts of the structure or meaning differ.
  • below 70% — possible drift into a neighboring intent; check page focus first.

If overall similarity is high but individual questions or entities are missing, improve the identified sections rather than rewriting the entire page.

How it is calculated

  • cleaned text is collected from your page and competitors;
  • an embedding vector is created for each document;
  • your vector is compared with each competitor using cosine similarity;
  • the average top-results vector and your similarity to it are calculated separately;
  • the closest competitor is identified by the highest similarity;
  • the result reflects overall topical focus, not exact word matches.
How do I read the page edit plan?

This is a prioritized SEO edit plan assembled from questions, entities, structure, page profiles and cosine similarity.

The table shows what to add and how to approach it: edit type, suggested block, sample H2/H3, page location and reason.

How to read this section

  • Edit type — add or rewrite a block, add FAQ, or remove irrelevant content.
  • Page location — where the new topic fits: after the service description, before the CTA, beside a form or in FAQ.
  • Suggested H2/H3 — a sample heading addressing the identified gap.
  • Action — a specific task for the editor, SEO specialist or website owner.
  • Reason — the top-results signal behind the recommendation.

The section turns analysis into a page-editing plan rather than leaving you with raw metrics.

How it is calculated

  • identified gaps are collected from entities, questions, structure, page profiles and cosine similarity;
  • each gap receives a high, medium or low priority;
  • question gaps become H2/H3 sections or FAQ answers;
  • structural gaps become tasks to add or rewrite blocks;
  • entity gaps become tasks to explain a concept through an example, process step, risk, price or selection criterion;
  • topic drift becomes a task to remove or separate unrelated cards from the main content.
Where do the reasons behind recommendations come from?

This section explains why edits were recommended. It shows the problem and its evidence: the analysis layer, signal status, competitor findings and why they matter.

Unlike the edit plan, the emphasis is on the reason rather than the task. Use it to check the system's reasoning and decide whether a recommendation applies.

Analysis layers used in explanations

  • Entity frequency — an important entity appears among competitors but is absent or weakly covered on your page.
  • Question coverage — top results answer a question your page does not address directly.
  • Structure — competitors have a semantic block that is missing or covered differently on your page.
  • Page focus — adjacent topics in the page profile weaken the main intent.
  • Document cosine similarity — the page's overall meaning differs from the average top-results model.

The Competitor signal column identifies the evidence: a question, page-profile excerpt, competitor sentence or similarity percentage.

How it is calculated

  • each gap retains its analysis layer: entities, questions, structure, focus or embeddings;
  • the status is recorded: matching meaning, partial, missing or different meaning;
  • the closest match on your page is calculated for questions;
  • entity counts, top-results median and competitor presence are calculated;
  • sections of your page profile are compared with competitor sections;
  • the explanation includes a competitor question, sentence, profile excerpt or similarity percentage.
What do coverage and occurrences mean?

Coverage — terms and concepts present on most competitor pages but missing from yours.

Occurrences — the frequency of terms already used: how strongly they are represented relative to a consistent market reference.

This helps develop fuller, more expert topic coverage rather than simply adding words.

The market profile also shows the core: signals consistently present on most competitors. Rare elements on your page are not automatically errors; review them manually because they may be useful differentiators.

How is your page profile compared with competitors?

A page profile is a concise semantic description after HTML cleaning. It identifies the page's purpose, how it works, the problem it solves, what users receive and the key sections present.

Your page vs competitors shows the semantic sections on your page alongside similar competitor sections. The competitor column summarizes top results rather than quoting one specific site.

Why use a page profile?

It reveals differences in meaning rather than wording. A How it works block may discuss benefits on your page while competitors explain steps, timelines, responsibilities and outcomes.

Partial or Different meaning does not necessarily mean a heading is missing. It means the block covers a different set of ideas from those typical of top results.

How it is calculated

  • Title, meta description, H1–H3 and substantial text blocks are collected from each page;
  • technical, cookie, modal, hidden and navigation noise is removed where possible;
  • text is grouped into semantic sections: what it is, how it works, what it analyzes, the problem it solves and what users receive;
  • competitor excerpts are gathered for each section;
  • sections found among competitors but not on your page are marked Missing;
  • when both have a section, semantic similarity is calculated between your section and the top-results summary.
Which top-results questions does your page answer?

Question coverage shows which questions appear among competitors and how well your page answers them semantically.

The service extracts questions from competitor headings, FAQ, lists and text blocks, then compares them with your page's questions and headings using semantic similarity.

Reading the statuses

  • Matching meaning — the question is answered well on your page.
  • Partial — related meaning is present, but the answer is indirect or incomplete.
  • Missing — the question appears in top results, but your page has no strong answer.

Important questions are better addressed in a dedicated H2/H3 or FAQ answer than buried in a general paragraph.

How it is calculated

  • questions are extracted from competitor headings, FAQ, lists, summary, dt and text blocks;
  • noisy or overly long wording is filtered out;
  • question-like wording is also extracted from your page;
  • embedding vectors are created for competitor questions and your page's questions;
  • the closest question or heading on your page is found for each top-results question;
  • similarity determines the status: Matching meaning, Partial or Missing.
Which entities matter to the market?

Entity frequency shows important concepts repeated among competitors: services, methods, features, prices, risks, selection criteria, brands or objects specific to the niche.

This is not a list of words to insert mechanically. Explain each entity's role with an example, process step, risk, condition of use or selection criterion.

How an entity enters the table

  • the term appears among competitors;
  • it has a meaningful median frequency in top results;
  • it is present across a sufficient share of competitors;
  • it is absent or covered significantly less on your page.

Strong missing signals deserve priority. Weak signals need manual review: they may represent a useful topic, a brand, a menu item or a feature of one competitor.

How it is calculated

  • competitor text is tokenized; Russian forms are lemmatized, while English forms are counted separately;
  • term frequency is calculated for each competitor;
  • the median frequency across top results is calculated for each entity;
  • presence measures the share of competitors using the entity;
  • your page's frequency is compared with the top-results median;
  • importance increases when the entity appears frequently across a substantial share of competitors but is absent or weakly covered on your page.
What does the Page structure section mean?
What does HTML-zone contribution show?

The section compares Title, H1, H2/H3, body text, lists and anchors with a consistent top-results model. Weight shows a zone's influence in the hybrid formula; Your page and Top-results reference show term coverage quality in that zone.

Zone length is compared using words and characters in cleaned text. The market range spans the 25th to 75th percentile of admitted competitors: the central half of the sample, not a required text length. Words and characters are counted after HTML cleaning but before stopword removal.

A low score is not a reason to add words mechanically. Check intent and page type first, then naturally develop missing ideas in the appropriate HTML zone.


What does Vocabulary outside top results mean?

Absent from competitors means terms missing from every reference page. Rare means terms found on fewer than 50% of those pages. Occurrences count every repetition; distinct terms use normalized forms where supported. Shares use meaningful terms after the selected cleaning filters. Each competitor is compared with the other admitted pages; your page is excluded from the reference. Comparison is unavailable when there are no other pages. Brands, cities and unique benefits can appear in these groups, so review meaning rather than automatically deleting text.


What is Relevant-layer volume?

The number of meaningful market-core terms actually used on the page.

It indicates how fully the page covers the niche's subject matter.

A value below the market reference suggests fewer topics and subtopics are covered.

What to do:

Add missing sections, features, scenarios and answers consistently found on relevant top-ranking pages.


What are Unique core terms?

The number of different core terms used on the page.

This reflects the breadth and variety of topic coverage.

What to do:

Rather than repeating the main phrase, cover related concepts, features, use cases and limitations.


What does the H2 count show?

The number of main semantic sections on the page.

What to do:

Divide the material into substantial sections appropriate to the intent and page type.


What does the H3 count mean?

The level of detail within main sections.

What to do:

Use subheadings within complex H2 sections when they cover distinct aspects of the topic.


How important are H4 headings?

H4 headings indicate more detailed subsections.

What to do:

Use H4 only for genuinely nested subtopics, not merely to style ordinary text.


What does H3 per H2 mean?

A measure of section detail.

What to do:

If H2 sections contain few subsections, add useful detail instead of leaving them superficial.


What does Semantic block count show?

The number of logical sections on the page.

What to do:

Add self-contained blocks needed for the specific query and page type.


What are Blocks containing the core?

The number of blocks that actually contain the niche's thematic core.

What to do:

Make sure main sections address the query's topic rather than consisting only of decorative or functional text.


What are Unique core terms per block?

A measure of topic coverage within each semantic block.

What to do:

Develop weak blocks with related concepts, features, examples and use cases.


What is Core concentration in one block?

The distribution of core terms across the page's structure.

It indicates whether the page relies on one strong section or develops the topic throughout.

What to do:

If the core is concentrated in the first block, distribute related ideas across suitable sections.


How can I use these metrics to improve a page?

The metrics show where a page trails the market: breadth, depth or structural organization.

What to do:

Improve topic coverage, detail and meaningful sections rather than increasing text length mechanically.

What does Top-results alignment mean in the HTML-zone table?

This metric shows how a term's use in a selected zone compares with the consistent competitor level. It is calculated separately for Title, H1, H2/H3, body text, lists and anchors.

Formula: your zone score / expected top-results score × 100%. The displayed value is capped at 100%, so exceeding the reference does not encourage unnecessary repetition.

  • 100% — the zone meets or exceeds the consistent top-results reference.
  • 80–99% — the term is represented close to the market level.
  • 1–79% — the term is present, but the zone is weaker than the typical level.
  • 0% — the term is absent from this HTML zone.

This is more than a count comparison. The hybrid model considers cleaned zone length, BM25-style frequency saturation, HTML-zone weight, competitor presence, page types and repeated domains. The expected level uses a weighted competitor median, reducing the effect of an overoptimized page.

For example, if a term's expected H1 score is 1.8 and your page scores 0.9, alignment is 50%. Improve the H1 itself rather than repeating the word throughout the text.

Why does a term have an Entity badge?

Some unigram rows have an Entity badge because the term matches an entity in the top-results semantic model.

When a unigram appears in the entity list, its row receives the label, category, competitor presence and importance.

This helps distinguish common frequent words from concepts relevant to the page's intent.

The Entity badge does not mean you should merely insert the word. Develop the concept with an example, explanation, process step, limitation, price, risk or selection criterion.

How it is calculated

  • the unigram table is built from normalized single-word terms;
  • the semantic model separately produces a top-results entity list;
  • a matching unigram receives an entity badge;
  • the badge includes entity category, competitor presence and importance;
  • terms outside the entity list remain ordinary unigrams without a badge.
Why do tables have percentage bars?

Percentage bars provide a quick visual indication of signal strength, making high, medium and low values easier to scan.

Bars represent different measures in different sections:

  • entities: the share of competitors using an entity;
  • questions: semantic similarity between a top-results question and your page;
  • documents: cosine similarity between the page and competitors or the top-results model.

Bars do not replace recommendations. Make decisions using the status, competitor signal, explanation and action plan together.

Should navigation menus be included in the analysis?

In most cases, exclude menus from the main analysis.

🎯 The problem with navigation

Navigation blocks (menu, header, footer):

  • repeat across the website
  • may appear twice for desktop and mobile
  • contain many unrelated sections
  • do not reflect the meaning of the specific page being analyzed

For example, the menu on an SEO services page may contain:

  • Web development
  • PPC advertising
  • SMM
  • Ad targeting
  • Blog
  • Case studies

These terms begin to affect:

  • topic coverage scores
  • term occurrences
  • n-gram weight distribution
  • the consistent core

But they describe site structure rather than page content.

🚨 How this distorts the model

Including the menu can:

  • dilute the thematic core
  • create misleading overlaps among competitors
  • artificially strengthen secondary topics
  • reduce the accuracy of term-frequency analysis

This is especially relevant for agencies and large websites with extensive navigation.

🟢 When including menus makes sense

  • analyzing site architecture
  • researching internal-linking strategy
  • modeling the overall business structure

For a single landing-page analysis, menus are usually noise.

🏆 How the service handles this

You can exclude navigation blocks from the analysis.

Analyzing the page content is recommended by default, since the service compares it against a consistent market model derived from competitors.

Key idea:
Menus describe site structure.
Content conveys the page's meaning.
The service analyzes that meaning.

What does Ignore hidden blocks do?

This option excludes content present in HTML but not visible after the page loads. It applies equally to your page and competitors.

How hidden content is detected

When analyzing a URL, the page is opened in a browser, which exposes HTML attributes and styles from linked CSS files. The service recognizes display: none, visibility: hidden, opacity: 0, the attributes hidden and aria-hidden="true", as well as closed tabs, accordions and popup blocks.

It checks the entire ancestor chain. If an outer container is hidden, all nested text, headings and links are also treated as hidden, even without their own hiding class or style.

Example: hiding through an external CSS file

/* custom-style.css */ .mydiv { display: none; } <div class="mydiv"> <p>Hidden page text</p> </div>
    

The browser applies the rule from custom-style.css, and the contents of .mydiv are excluded from the analysis.

Example: a hidden parent container

<div style="display: none"> <section> <h2>Hidden heading</h2> <p>Hidden service description</p> <a href="/service">Hidden link</a> </section> </div>
    

The elements section, h2, p and a have no hiding rule of their own, but are excluded because their parent is hidden.

What is excluded

  • words in hidden blocks do not contribute to TF-IDF or coverage recommendations;
  • hidden H1–H4 headings do not contribute to structure analysis;
  • hidden links do not enter the anchor zone;
  • fully hidden semantic sections are not counted as visible blocks;
  • the hidden-text ratio is still calculated for page diagnostics.

Tabs, accordions and Show more buttons

The service evaluates the page after loading. Closed tabs or accordions are treated as hidden; the active open tab is counted as visible. This allows comparison of immediately visible content separately from the page's full HTML.

When to enable it

  • comparing the main visible content of landing pages;
  • when menus, popups, tabs or accordions substantially inflate the text;
  • when you need a reference without technically hidden blocks.

When to leave it off

  • when evaluating all content available to users, including FAQ and tabs;
  • when hidden blocks are an important part of the page's content.

For manually pasted HTML, external CSS cannot be loaded. Hiding is detected through inline styles, HTML attributes and common classes. Provide a URL to check arbitrary external CSS rules fully.

Why use Minimum competitor presence?

The filter includes terms found on a selected share of competitors, such as 50%, 75% or 100%.

This isolates the consistent niche core and excludes incidental terms specific to individual sites.

How many competitors should I use?

The number of competitors affects how selective the semantic model becomes.

2 competitors — a broader model.
More terms enter the vocabulary, including less consistent and more variable elements.

7 competitors — a more selective model.
Only consistent terms repeated across the competitive sample are retained.

As more competitors enter the sample, the thematic core generally becomes narrower and more consistent, retaining the niche's structurally important elements.


Which should I choose?

  • To identify a consistent market core, use 7 competitors.
  • To collect more ideas, explore varied wording and identify more semantic directions, 2 competitors may be suitable.
Why does the term set shrink when I add competitors?

Adding competitors makes the model more selective.

The service builds a consistent niche model from shared terms and structural elements. A larger sample raises the requirement for a term to be consistently represented.

Terms found on only one or two competitors may be specific to those sites. As the sample grows, these elements are filtered out.

As a result:

  • there are fewer incidental terms;
  • the thematic field narrows;
  • the niche core becomes more consistent.

Put simply:

  • 2 competitors → broader model, more terms;
  • 7 competitors → more selective model, retaining the consistent core.

This is expected: the system filters inconsistent elements rather than losing terms.

Why does a 50% filter barely change the number of terms?

Minimum competitor presence does not affect the model linearly.

At 50% only terms found on at least half the competitors remain. In many niches, key terms already appear on two or three sites.

Switching from No filter to 50% may therefore leave the term count nearly unchanged because many terms already meet the condition.

Larger changes become apparent at 75% or 100%, when only terms consistently used by most competitors remain.


Put simply:

  • 50% — broad filtering, mainly removing incidental terms
  • 75% — isolates the niche core
  • 100% — retains only the most consistent elements

This is expected when the niche already has an established semantic structure.

Why are word forms combined in the table?

For Russian content, the analysis uses lemmatization — reducing words to their dictionary form. Russian inflected forms of “cleaning” are counted as one term: “cleaning”.

This avoids splitting the frequency of a single concept across its Russian word forms, which would distort TF-IDF and IDF statistics. English forms are currently counted separately.

Without lemmatization, different Russian grammatical forms of the same word would be counted as separate terms even though they express the same concept.

For Russian, morphological analysis reduces words to their dictionary forms so the analysis compares semantic unitsrather than surface word forms.

This produces a more consistent niche model and relevance calculation for Russian content. For English content, singular/plural and other word forms remain separate.

What is TF-IDF and how is it used in SEO?

TF-IDF — a statistical measure of a term's importance in one document relative to a collection. TF measures frequency on a page; IDF reduces the weight of terms common throughout the collection.

In SEO, TF-IDF helps compare your vocabulary with search-result pages, find missing topical terms and identify terms used substantially more or less than the market reference. Here it is combined with structure, coverage, occurrences, anchors and semantic analysis.

TF-IDF is not Google's or Yandex's ranking formula. It is a transparent diagnostic method for preparing an evidence-based content improvement plan.

Where is TF-IDF used today?

TF-IDF is used in information retrieval, document comparison and classification, topical clustering, recommendation systems and text processing. SEO applications include search-result analysis, content auditing and identifying vocabulary characteristic of a topic.

Modern search engines also use neural and semantic models, so TF-IDF alone cannot assess page quality. Its value is in understandable document comparisons and identifying specific terms, phrases and topic gaps.

How does TF-IDF affect content rankings?

A TF-IDF score does not guarantee ranking improvements and is not an independently confirmed ranking factor. The analysis can help improve topic coverage, close vocabulary gaps and align content more closely with search intent.

Rankings also depend on content quality and usefulness, query type, site structure, technical health, links, freshness and other signals. Use TF-IDF as one part of an audit, not a requirement to reproduce competitor frequencies.

How do I run a TF-IDF analysis?
  1. Enter the query, search engine and appropriate region.
  2. Check that your page and the top-ranking documents address the same intent.
  3. Exclude irrelevant sites and adjust menu and hidden-block cleaning if needed.
  4. Run the unigram, bigram and trigram analysis.
  5. Review document relevance, coverage, occurrences and anchor terms.
  6. Read competitor context examples and prepare an edit plan.

The service automates document collection, HTML cleaning, calculations and recommendations. Choosing the right query, region and comparable page types improves the usefulness of the results.

How do I use TF-IDF results?

Start with missing terms consistently found on relevant competitors. Add them only to suitable sections: descriptions, answers, features, headings or link anchors. Use report context examples to choose natural wording.

Then review occurrence recommendations: important terms may need more coverage, while excessive repetition may need reducing. Do not copy every competitor's frequency or insert term lists. Improve completeness and clarity while preserving natural language and the page's commercial or informational purpose.

After editing, rerun the analysis and compare coverage, occurrences and overall relevance with the previous result.

Start using AI SEO Expert Tools

Create expert content faster and analyze SEO content in greater depth.
Get 30 trial credits when you register.

🔐 Sign in to your account

Sign in to use this tool.

⛔ Your trial has ended

Your Basic trial plan has ended.

Access to tools is restricted until you renew or change your plan.

🚀 No credits remaining

You have used all your available credits.

⚠️ Check your information

Unable to complete the action.

Channel
Group
Support chat