See whether the heading hierarchy describes the search intent Heading Vector Similarity Analyzer measures the semantic relationship between a calibrated search phrase and every indexed H1 through H6 heading on a page. Headings are more than large text. They label the sections a visitor scans, define the structure assistive technology announces, and communicate how supporting ideas relate to the primary subject. The analyzer turns each heading into an embedding, compares it with the search-phrase embedding, and displays the result as its own attributable row. This is not an exact-keyword checklist. A heading can earn strong semantic similarity by expressing the same useful idea with different natural language. Conversely, a heading that contains one query word may still describe an unrelated section. Use the rows to understand meaning, not to force identical phrasing throughout the hierarchy. The goal is a page whose labels accurately preview the information beneath them and collectively support the reason the page exists.
Read heading influence and weighted contribution The Ranking Sensor gives higher-level headings more influence: H1 receives a weight of 100 percent, H2 90 percent, H3 80 percent, H4 70 percent, H5 60 percent, and H6 50 percent. For each row, direct cosine similarity is multiplied by that heading-level influence. The tool displays both numbers and the resulting contribution, such as 72 similarity points multiplied by 90 percent influence to produce 64.8 of 90 possible points. After scoring all eligible headings, the system adds their weighted contributions and divides by the total available influence. This weighted average prevents a page with many headings from winning merely through quantity, while preserving the structural importance of the H1 and upper-level sections. Expand the score calculation to see the numerator, denominator, and result. A weak H1 deserves attention before a similarly weak H6 because it defines more of the page’s primary semantic structure.
Distinguish headings from body passages and title signals This tool evaluates section labels, not the complete paragraphs below them. AI Overviews Visualizer remains focused on passage embeddings, clusters, and answer alignment. META Title Vector Similarity Analyzer evaluates one title. Heading Vector Similarity Analyzer owns the H1 through H6 factor and its hierarchy weights. Keeping these tools separate prevents a user from interpreting a passage heatmap as though it explained a heading score. A heading can align well while the following section is thin or off topic, so review the actual page before publishing. The reverse can also occur: a valuable section may have a generic label such as “Overview” that hides its relevance. In that case, rewriting the heading can improve scanability and semantic structure without changing the paragraph. Use Content Depth for useful breadth, Query Proximity for literal word spacing and order, and semantic passage tools when the concern lies inside the body copy.
Compare target and outperformer structures responsibly Authenticated comparison mode loads an outperformer already indexed by the selected customer’s Ranking Sensor. Both sides use catalog heading embeddings and the same calibrated-query vector. The showdown lists every target heading beside the comparison page’s own hierarchy, including tag, text, similarity, influence, and contribution. This reveals structural differences that one combined percentage would hide. Do not reproduce the outperformer’s outline without checking intent. A university program page, association directory, and editorial guide may rank for the same phrase while answering different needs. Look for useful principles: Does the comparison page use a specific H1? Do its H2 labels expose decisions or requirements the target buries? Does the target contain repeated generic headings? Preserve original information architecture and add a section only when its content belongs on the page.
Test a revised hierarchy before changing the website The editable heading field uses one line per heading in the format “H1 | Heading text.” Keep the H1 through H6 tag at the beginning so the preview can apply the correct influence. Change one or more lines and choose Measure This Change. Fresh embeddings are generated for the edited labels, and every semantic row is rebuilt. The preview score and weighted formula therefore respond to the actual proposed wording rather than a made-up lexical substitute. Use Restore Indexed Content to return to the catalog hierarchy. If a line is omitted or does not use a valid H1 through H6 prefix, it cannot contribute to the edited preview. That validation is intentional because the factor measures headings, not arbitrary sentences. Keep the order consistent with the rendered document, retain only headings that truly exist, and avoid converting ordinary text into a heading solely to increase the number of rows.
Write headings that help visitors take action A strong heading states what the next section will resolve. Replace labels such as “More Information,” “Details,” or “Resources” with specific language when the section has a clear purpose. For a training page, headings might identify admission requirements, clinical experience, accreditation, program length, or career outcomes. For a product page, they might identify fit, capabilities, setup, limitations, proof, or pricing. Choose only subjects supported by real content. Do not turn every heading into a long sentence or repeat the calibrated phrase six times. Repetition can make the outline harder to scan and blur distinctions between sections. Related supporting concepts often produce good semantic alignment naturally. Read the headings alone as an outline. A visitor should understand the page’s progression, identify the section relevant to their question, and predict what evidence or next step each label introduces.
Respect accessible document structure Heading tags carry structural meaning. Maintain a logical hierarchy rather than selecting a level for font size. A page generally has one primary H1, followed by H2 sections and nested H3 or lower subsections when needed. The analyzer’s weights reflect importance but are not permission to promote every label to H1. Doing so would damage accessibility, scanability, and document organization even if a numeric preview appeared favorable. After editing, inspect the rendered page with styles enabled and, when possible, review its heading outline or accessibility tree. Confirm that no navigation, modal, footer, or hidden component injects misleading headings into the indexed set. Template headings may appear across many URLs and dilute the page-specific structure. Correct the component at its source when the same generic label causes repeated problems rather than patching individual pages.
Recalculate the Ranking Sensor after publication The weighted cosine average is the Headings Similarity score shown in both the task and analyzer. Ranking Sensor calibration controls how strongly that measurement influences modeled rankings without replacing it with another percentage. Once the approved hierarchy is live and recrawled, recalculate the sensor to generate new stored embeddings and determine whether the measured task gap closed. If the live score differs from the preview, verify which headings were actually indexed. Client-side rendering, duplicate mobile markup, hidden accordions, template changes, and crawler timing can alter the stored set. Also remember that changing headings can affect other signals and user behavior. Review the corresponding body sections, title, navigation, and conversion path. Treat vector similarity as transparent evidence supporting a clearer hierarchy, not as an isolated target that overrides accuracy or usability.
