On-Page Optimisation
Run on-page optimization for titles, descriptions, headings, entities, anchors, and media, with measurable checks for new and existing pages before launch.
On-page optimization is the page-level alignment of titles, descriptions, headings, entities, internal anchors, and media with one useful reader task. The post type already decides the page’s structural job; this phase makes every visible and machine-readable signal describe that job consistently.
Phase: P11 · Stage C — Produce. Timebox: 60–90 minutes for a new page after editorial approval; two to four hours for an existing page because diagnosis, preservation, and before-and-after evidence are required. Accountable owner: SEO content lead. Contributors: writer or editor, SEO strategist, designer for material media changes, and developer when templates emit the wrong HTML.
Why this phase, and why here
P11 consumes the approved draft and production controls from the content production system , plus the page’s intent, canonical URL, primary entities, required evidence, and post-type specification . Structure has already answered whether the asset is a guide, comparison, product page, glossary term, or another format. On-page work should not redesign that decision by squeezing every query variant into the page.
The phase exists because the same accurate article can send conflicting signals. A browser tab might promise “Enterprise CRM Migration Guide,” the H1 might say “Moving Your Data,” the introduction might never name the systems involved, and internal links might call the page “read more.” A person can infer the relationship after reading. A search engine, screen reader, or AI retrieval system must reconcile several weaker labels before it can decide what the page is about and when it is useful.
The order matters. Run on-page optimization before the dedicated internal linking phase because P12 needs the final URL, stable headings, and approved anchor concepts. Run it after drafting because a title written before the answer exists often promises a scope the body does not deliver. Skipping P11 leaves useful content harder to classify and less compelling to click. Running it during research encourages keyword-led copy before search intent —the task a searcher is trying to complete—has been settled.
Inputs and outputs
Outputs are the contract with internal linking and pre-publish quality assurance. An editor should not need to infer what the title meant or which heading can safely receive a deep link.
| Direction | Item | Acceptance condition |
|---|---|---|
| Input | Approved page brief and post type | Names one primary intent, audience, page job, required elements, and conversion action. |
| Input | Editorially approved body | Contains the complete answer, supporting evidence, and no unresolved factual placeholders. |
| Input | Entity and terminology list | Names the products, organizations, people, locations, standards, and preferred spellings that matter to the answer. |
| Input | Existing-page baseline, when applicable | Records query, click, impression, CTR, position, conversion, and current snippet evidence for a fixed date range. |
| Input | Technical publishing constraints | Confirms the canonical URL, indexability, template fields, media budget, and who can change rendered HTML. |
| Output | Approved snippet set | Supplies one title, one description, and one H1 whose promise agrees with the visible answer. |
| Output | Machine-readable outline | Supplies a logical H1–H3 hierarchy with stable, descriptive section labels. |
| Output | Entity and anchor map | Records the preferred entity names, clarifying statements, target links, and natural anchor concepts used on the page. |
| Output | Media manifest | Lists every meaningful asset, purpose, filename, dimensions, caption need, alternative text, and performance status. |
| Output | On-page change record | Preserves before and after values, reason, owner, publication date, acceptance result, and measurement date. |
Ready to Monitor Your AI Visibility?
Track how AI chatbots mention your brand across ChatGPT, Perplexity, and other platforms.
The checklist
The checklist differs at the first step for an existing page and a new page. The remaining checks share acceptance standards, but evidence must never be discarded simply to make an old page resemble a new template.
1. Choose the existing-page or new-page route
What: Classify the work as existing-page optimization or new-page completion before changing copy. An existing page has measurable history and may already satisfy queries outside the current brief; a new page has no performance baseline to protect.
Why: Editing a live performer without capturing its working coverage makes losses impossible to diagnose. Treating a new page like a declining page invites made-up baselines and premature success claims.
How: For an existing page, export the current title, description, H1–H3 outline, internal anchors, media, top queries, clicks, impressions, CTR, average position, and conversions for a declared range. Annotate what must be preserved. For a new page, verify the approved intent, post type, URL, entity list, and conversion action; mark performance fields “baseline after launch,” not zero.
Tool: Use Google Search Queries at app.amicited.com/reports/google-search/queries for live query evidence. Use the CMS preview and the approved brief for a new page.
Done when: The change record states the route, evidence window, owner, protected queries or sections, and reason for the work. No existing-page text changes begin without a captured baseline, and no new page is judged against fabricated historical performance.
2. Reconfirm the page’s single primary promise
What: Write one sentence: “This page helps [audience] complete [task] by providing [answer or decision support].” Mark one primary query theme and the important secondary questions the body genuinely answers.
Why: Titles, headings, entities, anchors, and media can only agree when the page has one dominant job. A list of keywords is not a promise because it says nothing about the outcome the reader receives.
How: Compare the approved intent with the opening answer, evidence, result-page expectations, and call to action. If the draft serves two different tasks with different evidence or next actions, return it to scope review rather than hiding the conflict inside a broad title.
Tool: Use the brief, live result review, and the query evidence attached in step 1.
Done when: A reviewer can read the promise and point to the direct answer, supporting sections, and next action that fulfill it. Every retained query theme fits that same job.
3. Write a title that can survive rewriting
What: Finalize the HTML title tag —the page name shown in a browser tab and commonly used as the search-result headline—and its visible H1.
Why: Search engines may rewrite titles when they are repetitive, vague, stuffed, stale, or inconsistent with the visible page. No wording can prevent every rewrite, because results adapt to queries and devices. A specific, concise title that agrees with the H1 and answer gives the system less reason to replace the page’s framing.
How: Lead with the subject and useful outcome, add a differentiator only when the body proves it, and put the brand last when it helps identification. Remove boilerplate shared across hundreds of pages. Keep the H1 natural and slightly more readable than the title if needed, but make both describe the same scope. Preview width rather than treating a character count as a ranking rule.
Tool: Use the CMS search preview, live result comparison, and Google Search Queries for the language people actually use.
Done when: The rendered source contains one unique, non-empty title and one H1; neither is a list of keyword variants; their promises agree with each other and the opening answer; and the title’s important words remain understandable if its tail is truncated.
4. Write a description that earns the right click
What: Write the meta description , an HTML summary that search engines may show beneath the result title.
Why: The description is not a guaranteed snippet and is not a place to force rankings. Its job is to make the page’s value and qualification clear so the right searcher can choose it. Search engines commonly select visible page text when that text answers a query more precisely.
How: State the subject, outcome, useful constraint, and next step in plain language. Use the primary term where natural. Aim for 120–160 characters as an editorial range, then preview desktop and mobile. Do not repeat the title, use unsupported superlatives, or promise tools, prices, templates, or evidence absent from the page.
Tool: Use the CMS preview and CTR Gap at app.amicited.com/reports/ctr-gap for existing results earning fewer clicks than the site’s fitted expectation.
Done when: The description is unique, accurate without context, readable in the preview, and supported by the page. For an existing CTR-gap edit, the previous description, diagnosed cause, replacement, and review date are recorded.
5. Turn headings into the page outline
What: Make the H1, H2s, and H3s a nested outline of the answer. Heading levels are semantic HTML labels, not controls for font size.
Why: Readers scan headings to decide where to invest attention. Assistive technology uses them for navigation, while search and retrieval systems use them to associate passages with questions and entities. Decorative or empty headings corrupt that outline.
How: Use one H1 for the page. Give each major section an H2 and reserve H3 for a real subdivision of its parent. Rewrite vague labels such as “Overview,” “More,” and “Benefits” so they name the subject in context. Move styling needs to the design system; do not select H4 because it looks smaller. Every heading must introduce text, a table, a list, media, or another substantive answer.
Tool: Use the rendered DOM outline, not only the visual editor. Check both desktop and mobile previews.
Done when: There is exactly one H1; no level is skipped merely for appearance; every H3 belongs to the preceding H2; no heading is empty or duplicated without a clear repeated structure; and reading headings alone gives a truthful summary of the page.
6. Name entities and state their relationships
What: Verify the page’s entities—distinct people, organizations, products, places, standards, methods, or measurements—and the relationships asserted between them.
Why: Repeating a keyword does not resolve whether “Mercury” means a planet, an element, a car brand, or a payment company. Clear names, categories, attributes, and relationships help a reader and a machine connect claims to the right thing.
How: Use the preferred full name on first mention, define specialized terms, and state important relationships in complete sentences. Add versions, locations, dates, units, and authorship where they change meaning. Use synonyms naturally after disambiguation. Remove entity lists that have no explanatory relationship and verify every factual association against the approved source material.
Tool: Use the entity list from the brief, editorial source log, and rendered page search. AmICited query evidence can show the vocabulary users apply, but it does not verify factual relationships.
Done when: Every primary entity is unambiguous on first meaningful mention, every material relationship has evidence, names and versions are consistent, and an editor can extract an entity–relationship list without guessing what a pronoun or acronym refers to.
7. Make internal anchors describe the next useful step
What: Review the anchor text —the visible clickable words—for every internal link already assigned to the page.
Why: “Click here” and “learn more” hide the destination from people scanning the page and from systems interpreting the relationship. Exact-match repetition is not the cure; anchors should describe why the destination helps at that point in the answer.
How: Place links where the destination resolves a question, provides proof, or enables the next task. Use concise descriptive language that fits the sentence. Vary wording when context changes, avoid linking the same destination repeatedly within one short section, and never add a link solely to place a target phrase.
Tool: Use the approved link obligations, the CMS link inspector, and the rendered page. The next phase will evaluate graph coverage and source-page opportunities.
Done when: Every internal link resolves to the intended canonical URL, zero generic “click here” or bare-URL anchors remain, the destination is understandable from its sentence, and the handoff lists the stable headings and concepts P12 may use for inbound links.
8. Optimize media for meaning, access, and speed
What: Review images, diagrams, charts, video, and embeds for purpose, placement, dimensions, file format, captions, and alt text —the textual alternative announced when an image cannot be seen.
Why: Media should explain something prose cannot show as efficiently. Unlabelled charts hide evidence, missing dimensions cause layout shifts, and decorative files with verbose alternative text create noise for screen-reader users. Large assets can make the answer slower without making it clearer.
How: Keep each asset only when it proves, explains, or demonstrates a point. Write concise alt text for meaningful images based on their function in context; use empty alt text for purely decorative images. Put trends and takeaways from charts in visible prose, supply width and height, use an efficient format, and lazy-load below-the-fold media where the implementation supports it. Capture product screenshots at a legible viewport and redact personal or customer data.
Tool: Use the media manifest, browser accessibility inspection, image-dimension check, and the site’s agreed performance budget.
Done when: Every media item has an owner and purpose; informative assets have suitable alternative text; decorative assets use empty alternatives; charts state their takeaway in text; dimensions are declared; no sensitive data is visible; and every file passes the site’s media budget or has an approved exception.
9. Render, compare, and approve the complete page
What: Review the rendered page as one system and record the change set.
Why: Fields that pass alone can conflict together. A concise title may narrow the scope while an old H2 broadens it; a strong description may promise a template removed in editing; a new heading ID may break an inbound deep link.
How: Compare the title, description, H1, opening answer, outline, entities, anchors, media, and CTA against the primary promise. Inspect HTML output, desktop, mobile, keyboard navigation, and the live destination of every link. For an existing page, separate changes by hypothesis so later measurement can identify what likely moved.
Tool: Use the CMS preview, browser inspector, link checker available to the publishing team, and the AmICited baseline captures.
Done when: The acceptance gates below all pass, the accountable owner approves the rendered URL, before-and-after evidence is attached, the publication and measurement dates are set, and unresolved issues have an owner rather than disappearing into a comment.
Tools in AmICited
AmICited identifies pages worth changing and supplies evidence for the change. It does not replace review of the live result, rendered HTML, or page promise.
| Product view | Use in this phase | Deep link | Evidence to retain |
|---|---|---|---|
| Google Search Queries | Identify the query language, demand, clicks, CTR, and position attached to an existing page before editing. | Open the Queries report | Date range, filters, query rows, affected URL, and export date. |
| CTR Gap | Find queries or pages earning fewer clicks than this site’s own fitted CTR curve predicts, then prioritize a title, description, or intent diagnosis. | Open the CTR Gap report | Expected CTR, actual CTR, clicks at stake, fitting level, live-result diagnosis, and proposed remedy. |
| Striking Distance | Group queries near the chosen target band by the page that owns them, so one coherent page improvement can support the cluster. | Open the Striking Distance report | Position band, target position, minimum impressions, owning page, qualifying queries, and modelled upside. |
Stay Updated on AI Visibility Trends
Get the latest insights on AI mentions, brand monitoring, and optimization strategies.
Decision rules
These numbers are review gates, not universal algorithm thresholds. A documented exception can pass; an invisible exception cannot.
| Check | Bad looks like, in numbers | Required action |
|---|---|---|
| Page-level labels | Title count is not 1, H1 count is not 1, or either field is empty. | Block publication until the rendered HTML has one of each. |
| Duplicate title | 2 or more indexable URLs use the same full title without a deliberate series convention. | Differentiate the page job or resolve the underlying overlap. |
| Description | It is missing, duplicated, below 90 characters, or above 180 characters without an editorial reason. | Rewrite toward the 120–160 character working range and verify the promise. |
| Heading support | A heading has 0 substantive content blocks before the next heading at the same or higher level. | Add the promised answer or remove the heading. |
| Outline depth | A heading skips from H1 to H3, or an H3 has no H2 parent. | Repair the semantic hierarchy; change styling separately. |
| Generic internal anchors | 1 or more anchors use only “click here,” “here,” “read more,” or a bare URL. | Replace with destination-and-purpose language. |
| Media accessibility | 1 or more informative images lack alternative text, or decorative images announce filenames. | Supply functional alt text or an empty alternative as appropriate. |
| Existing-page evidence | 0 baseline captures or 0 declared comparison windows exist before the edit. | Pause the edit and capture what must be protected and measured. |
| CTR diagnosis | A page is below 75% of fitted expected CTR—the red band in CTR Gap—but no live-result inspection is recorded. | Inspect intent, SERP features, and competing snippets before prescribing copy. |
| Striking-distance scope | Queries fall outside the team’s declared position band or have 0 impressions in the evidence window. | Exclude them from the optimization hypothesis; do not pad the target list. |
| Change isolation | More than 3 material dimensions change with no reason or annotation. | Split the release where practical or record why combined change is necessary. |
Do not use keyword density as a gate. Density is a ratio of phrase occurrences to total words, but it cannot tell whether a page answers the question, distinguishes entities, or reads naturally. Zero forced insertions is the standard. Likewise, an exact-match heading with no content behind it fails even if a tool marks the phrase as “optimized.”
Deliverable: the on-page change sheet
Hand over one versioned row per URL, linked to the CMS-ready copy and evidence folder.
URL | Route: existing/new | Primary promise | Audience | Post type
Title before | Title after | Description before | Description after | H1
Heading outline | Primary entities | Internal anchors | Media manifest
Protected queries/sections | AmICited evidence links | Change hypothesis
Owner | Reviewer | Publish date | Measurement date | Status | Exceptions
For existing pages, include before-and-after proof and exported query evidence. For new pages, include the approved brief and mark performance fields “awaiting baseline.” Record every exception beside its failed gate. Publishing must be able to implement the sheet without rewriting fields, and measurement must be able to reconstruct the change.
What goes wrong
- A score replaces judgment. A plugin turns green because a phrase appears often enough, while the page answers the wrong task. Return to the primary promise and observed query intent.
- The snippet overpromises. A keyword-filled title or an unsupported promise such as “free template” earns the wrong click and invites rewriting. Keep one subject, one outcome, and only claims the body fulfills.
- Headings are decoration. Editors choose heading levels for size, or add question headings followed by one empty sentence. Repair the outline in HTML and use design styles for appearance.
- Exact-match anchors multiply. Every link to a commercial page uses the same awkward phrase. Write anchors for the local sentence and user need; consistency of destination does not require identical wording.
- An existing winner is rewritten like a blank page. Useful subtopics and language disappear because the new brief only records the primary query. Preserve adjacent coverage and log deliberate removals.
- Every low CTR becomes a copy problem. An AI overview, image pack, brand mismatch, or wrong landing page may suppress clicks. Inspect the live result before changing the snippet.
- Search-engine rewriting triggers daily edits. One observed title variation causes reactive changes that erase the experiment. Collect repeated, query-specific evidence and change only when the page itself is misaligned.
Next phase
The internal linking phase receives the final canonical URL, primary promise, stable heading outline, approved entities, existing outbound links, and candidate anchor concepts. It uses those fields to decide which relevant pages should link in, which contextual routes should lead out, and how the page fits the site’s wider graph.
Do not hand off a provisional heading map or a URL that may change after publication. P12 should decide link placement and coverage, not reopen the page’s intent or invent labels for an unfinished answer. The on-page owner remains responsible for any wording changes needed to make a planned link natural.
Put the page-level signals into agreement
Start with the page’s actual evidence: open the Queries report for the owning URL, use the CTR opportunity report when clicks underperform the fitted expectation, or open the page opportunity worklist when a cluster is close enough to improve. Then ship one documented page-level hypothesis, not a bundle of keyword insertions.
More tutorials in this section
Ready to put it into practice?
Free check · 7-day trial · no credit card