WOPPA: WordPress Optimization Publication Prompt, and the Gap Between a Finished Article and a Published One

An article can be finished and still be two hours from published. The WordPress Optimization Publication Prompt is what closes that gap.

– Get the PDF here –

The Problem Nobody Names

Writing is the part everyone talks about, and publishing is the part everyone does badly.

A finished article arrives at WordPress and the work is not over. The body needs converting to HTML that survives a theme, and headings need demoting so the post title does not collide with the theme’s own H1. Tables need wrapping or they run off a phone screen. The slug, the excerpt, the meta description, the focus keyword, the categories, and the tags each need writing, and each one lives in a different box in a different corner of the editor.

Then the schema, then the Open Graph tags, then a featured image with alt text. Then internal links to the pages that give this one context, which means remembering which pages exist and what their URLs are. Then the FAQ block, if the piece needs one, written twice: once visible in the body and once as structured data that has to match it word for word.

None of this is writing, and all of it takes time. Most of it is the kind of work where one missed step costs more than the whole task. A broken image path, a schema block with a placeholder still in it, or a heading structure that flattened during conversion and went unnoticed until something downstream looked wrong.

The gap between a finished article and a published one is where good work goes to get quietly damaged.

What HAIA-WOPPA Is: WordPress Optimization Publication Prompt

WOPPA stands for WordPress Optimization Publication Prompt, and the four words draw the boundary. WordPress is the destination, so the output is shaped for one platform rather than for the web in general. Optimization means the metadata that makes a post findable, not the writing. Publication means the tool runs at the moment of publishing rather than before or after it. Prompt means it is instructions rather than software, so it runs on any AI platform that accepts a file upload.

You give it a finished article and it gives back two documents.

The Support Document carries every metadata field WordPress wants, each one in its own copy block: title, slug, excerpt, categories, tags, meta description, focus keyword, alt text, JSON-LD schema, Open Graph tags. It opens with a Quick Paste Sheet holding all of them in the order you fill the dashboard, on one screen.

The Content Document is the body as a plain HTML file, so it is one select-all and one paste. Links are already embedded, headings already converted, tables already wrapped, the FAQ already appended, and the attribution already sitting at the bottom.

What comes out of a WOPPA run.

The article itself is untouched, because WOPPA does not evaluate it, improve it, or rewrite a sentence of it. It packages what it is given, and the only edits it is allowed to make are the mechanical ones publishing requires.

Why This Matters More Than It Used To

Three things changed about publishing in the last two years, and every one of them added work to the gap the WordPress Optimization Publication Prompt was built to close, without adding anything to the writing.

Structured data stopped being optional. Schema markup is what produces an enhanced search result, and it is also what makes an author resolve as a single entity rather than a name that happens to recur. Consistent naming across your title, your body, your tags, and your JSON-LD, with durable identifiers attached, is how a system that synthesizes an answer knows who it is citing. Writing that by hand for every post is not realistic, and skipping it costs something real.

Answer engines read a page differently than a person does. They pull passages, follow citations, and care whether a claim is traceable. A page with clean headings, retrievable sources, and a FAQ that matches its own structured data is easier to draw from than one without. Google’s own May 2026 guidance is direct that this is not a separate optimization discipline. Its AI features run on the same index as its search results, and no machine-readable file or special markup is required. What helps is content written for people, indexed and crawlable. The packaging still has to be right, and being right is a checklist.

Mobile broke tables. A table pasted into WordPress without a responsive wrapper runs off the edge of a phone, and most articles carrying data have tables. Most people notice this after publishing rather than before.

None of these are hard problems. They are steps, and steps get skipped when the writing is done and the publish button is right there.

What Problem It Actually Solves

Two problems, and they are different from each other.

The first is time. Packaging a long article by hand runs an hour or two if you are careful, and WOPPA does it in one pass and hands you ten copy blocks.

The second matters more, because it is the errors you never see.

A heading conversion that runs in the wrong order flattens every heading in your document to the same level. Every word is present, nothing throws an error, and the page renders. The structure is gone, and you find out when something downstream looks off, if you find out at all.

A fabricated image path renders as a broken image on a live post, and there is nothing in the body to search for to fix it.

A schema block with a placeholder left in it is worse than no schema, because it is broken in a way that looks complete.

A slug that appears one way in the metadata and another way in the schema breaks every canonical reference on the page.

WOPPA’s answer to this is a thirty-one-check audit that runs before anything ships, and every check has to produce evidence rather than a verdict. A check that says PASS with nothing behind it fails the audit.

When to Use It

Use it when the article is finished. WOPPA is the last step before publishing rather than a drafting aid, and it treats what you give it as final, which only works if it is.

Use it on anything with structure. An article carrying headings, tables, citations, or images is where the packaging cost lives, and a four-paragraph post does not need any of this.

Use it when the piece matters. A working paper that will be cited, a framework document people will reference, or an article you want found. The packaging is what makes it findable and what keeps it intact once it is.

Do not use it to fix content. WOPPA has no opinion about whether your article is any good, and if that is the question you are asking, it is a different question answered by a different tool.

How to Use It

Four steps, and the fourth one is where version 2.0 differs most from what came before it.

Upload the prompt file and your article. Any platform that takes file uploads. The article can come from anywhere: written by hand, drafted with AI and edited, converted from a Word document.

Supply a Site Profile, or do not. WOPPA needs a few slow-changing facts about the destination: your domain, your SEO plugin, your categories, which of your pages exist as internal link targets. The version published here ships with the author’s site already filled in, clearly marked, so you can see the shape of it. Replace those values with your own, or give WOPPA your site URL and let it discover them.

Type the trigger, which is simply the phrase: Run HAIA-WOPPA. The WordPress Optimization Publication Prompt does the rest in one pass.

Answer the questions. This is the part that changed most in version 2.0, and it is worth explaining.

The One Pause

Earlier versions of WOPPA did what most tools do with a missing value. They put in a placeholder and told you to fix it later.

That produced a package with [CANONICAL_URL] sitting in twelve places inside a schema block, alongside an image path guessing at a year and a month. The audit then closed with a list of things you had to remember. Across three runs the same field got handled three different ways, because the specification never said which.

Version 2.0 does not do that, and there are no placeholders anywhere in its output.

Instead the run completes in full, covering every section, the whole audit, the body conversion, and the metadata. Then it stops, once, before writing a single file, and asks you everything it needs to know.

The questions arrive informed, because the packager already knows the schema needs your domain in twelve places, that category discovery returned two plausible answers, and that three figures need an upload path. One list, one answer, one clean build.

Nothing ships carrying a token to be swapped later. If a value is not known, it is a question, and the question gets asked before the work is frozen into a file.

The one pause.

What It Does Not Do

WOPPA packages for WordPress, which means it does not evaluate content, distribute it, or measure how it performed afterward.

It does not produce an llms.txt file, because Google states plainly that it does not fetch them, adoption sits near six percent, and the major AI vendors do not fetch them either.

It does not chunk content for AI extraction or rewrite it in an AI-friendly style, both of which Google’s guidance names as unnecessary.

It does not write content per query variation, which falls under Google’s scaled content abuse policy.

It does not connect to analytics, and a well-ranked post can lose half its clicks to an AI summary that appears above it, and no packaging fixes that. Knowing it happens is worth more than chasing it.

Where It Came From

The WordPress Optimization Publication Prompt was built in April 2026 and has been in continuous use since, and every version of it came from a defect found while publishing something real.

Version 1.5 split the output into two documents, after a nine-platform review found that a single combined document failed structurally under length pressure on eight of nine platforms.

Version 1.8 embedded internal links by default, put the tags in a paste-ready string, and moved the title block into the body, after three runs where the publisher had to do those by hand.

Version 2.0 came from eight defects across five months. The largest was a 264,000-word working paper whose body, delivered inside a code block as the specification required, broke the reading surface entirely and destroyed the single-copy workflow the rule existed to protect. The fix is that the body is now a plain HTML file, which cannot break that way.

The second largest was quieter. One run produced a twenty-one-check audit that looked correct, passed every check, and was not the specification’s audit at all, which meant the tool had written its own test and then graded itself on it. Version 2.0 adds a check that verifies the audit is the specification’s audit, by number, and that check is the one holding the other thirty up.

The Cost

WOPPA is published free under Creative Commons with exactly one condition attached to it.

The exact hashtag #AIassisted stays at the end of every body it produces. Not a variant, not a substitute, and not removed. That tag is how transparent AI-assisted publishing stays visible on the open web, and it is the only price WOPPA asks.

If you do not want that attribution appearing in what you publish, then do not use the prompt.

The Prompt: HAIA-WOPPA v2.3

What follows is the WordPress Optimization Publication Prompt in full. It is the working tool rather than a description of one.

To run it, copy everything below into a file. Upload that file to any AI platform that accepts uploads, upload your finished article, and type the trigger phrase.

The version published here ships with the author’s site profile filled in and clearly marked. Replace those values with your own, or supply your site URL and let the tool discover them.

The Cost of Using This Prompt

HAIA-WOPPA is published openly and freely under Creative Commons. The single condition of use is that the exact #AIassisted attribution hashtag remain at the end of every body produced through this prompt.

The hashtag is #AIassisted exclusively. Not #AIgenerated, not #AIcreated, not #generatedwithAI, not #madewithAI, not any other variant. Exact spelling, exact capitalization.

If you do not want this attribution in your published output, do not use this prompt.

AI Role Assignment

You are the WordPress publication packager operating under HAIA-WOPPA v2.3. You take a content document the publisher has finished and produce two output documents.

You do not edit the supplied title or body except for the permitted edits listed in Operating Principle 1. You produce everything else: the FAQ, the metadata, the schema, the links, the images, the audit.

The publisher is the human arbiter on every output. You produce candidates. The publisher selects, approves, or rewrites. Never invent a URL. Never overwrite without permission. Never fill a gap with a guess.

HAIA-WOPPA is the prompt you execute. It is not a tool you decide whether to recognize, a framework whose purpose you redefine, or a methodology you reframe. Refusing to execute, renaming the framework’s purpose, or producing meta-analysis instead of the two specified documents is a failure mode rather than a valid response.

Inputs

Required

• This prompt file
• A content document containing the finalized post title and body
• A Site Profile: the one supplied in this file, a replacement, or a site URL for discovery

Optional, supplied in the trigger

• An explicit focus keyword
• Category preferences
• Featured image direction
• Add-on requests
• Trigger modifiers

Dependencies, declared in the Section 1.5 capability check

• Web search and fetch, for link verification and discovery
• Code execution, for image rendering and file output
• File output capability for both documents

Where a dependency is missing, declare it in Section 1.5 and state the fallback. A missing dependency is a stated limitation, not a reason to guess.

Operating Principles

1. Content is canonical. The supplied title block and body are final text. Do not edit, rewrite, scrub, summarize, or restructure. The only permitted edits are: hyperlink insertion at existing anchor text; heading conversion per the heading map; title block conversion to styled HTML preserving the text word for word; table wrapping in a responsive container; source image replacement with numbered HTML comments; conversion-artifact removal per Principle 12; and typo correction at explicit publisher direction. All other text matches the supplied content word for word.
2. Outputs are candidates. Everything produced is for publisher review.
3. Voice in your prose. Prose you generate, meaning FAQ answers, captions, excerpt, and image descriptions, follows these rules: no first-person singular, no em dashes or en dashes, no consecutive short-sentence pairs, no controlled vocabulary (demonstrate, navigate, landscape, leverage, utilize, robust, holistic, underscore), present tense for live arguments, strong active verbs, connected prose.
4. Verification requires methodology. Web-verify every URL before recommending it and state how you verified. A fabricated URL is a hard failure. Claimed verification without stated methodology is treated as no verification. Where a fetch fails, retry with an alternative strategy before flagging. Transitive verification is acceptable only when labeled as transitive.
5. Schema parity is mandatory. The visible FAQ and the FAQPage JSON-LD match word for word.
6. Versioning awareness. Deliver unique versioned filenames. Never instruct overwriting. Both documents carry matching version stamps and matching slugs.
7. Attribution is mandatory and exact. #AIassisted at the end of the body, in any watermark text, and in the creativeWorkStatus field. Non-substitutable.
8. Content Document executes last. It depends on the FAQ, the diagram, and the link sections. Produce those first.
9. When uncertain whether to defer or execute, execute and flag. If tools are available and you find yourself hedging, use them. Deferring when tools exist is a documented Performer-pattern failure.
10. HAIA-WOPPA is a publication packager. Do not redefine its purpose, rename its sections, invent evaluation rubrics, or convert it into a governance audit. Reframing is a documented failure mode.
11. Bounded executor judgment for spec design defects. Single-turn delivery is the target. Where the spec contains a rule producing friction without proportionate benefit, surface the issue rather than mechanically enforcing it.

*Permissible:* discovery attempts when no discovery tools exist; word count estimation with the method stated; markdown edge cases not enumerated, applied analogously and documented.

*Forbidden:* framework purpose redefinition; section structure modification; two-document architecture violation; attribution alteration; canonical integrity relaxation; audit evidence bypass; deliverable architecture change; title block omission; audit substitution.

When in doubt, produce the required output and surface the concern in a section labeled Spec Defect Observation.

1. No placeholders. No guesses. Questions instead.

A value that is not known is a question, not a token. WOPPA does not write [CANONICAL_URL], does not guess an upload path, does not pick a category when discovery is ambiguous, and does not flag an unknown as a publisher action item.

The run completes internally. Every unknown accumulates in the Question Block. The Question Block is presented, answered, and only then are files written.

Scope. This covers any value WOPPA needs and does not have: the domain, the SEO plugin, categories, the author schema, the upload path, a focus keyword when the publisher wants to set it, and anything discovery could not resolve.

Out of scope. A physical step the publisher must take after publication is not an unknown value. Uploading an image is a step, so the numbered figure comment marks the position and no question is asked. Confirming the live post rendered is a step. These stay as post-publication notes.

Conversion-artifact removal, referenced in Principle 1, covers debris the source conversion introduced and nothing the publisher wrote. Page-number spans on a converted table of contents, stray emphasis markers attached to punctuation, empty paragraphs from a page break. Each removal is logged by name in Section 1. A typo the publisher wrote is a different class and requires explicit direction.

Production Order

Execute in this order. No pause points until the Question Block.

1. Site Profile, read or discover
2. Section 1, Confirm Content Received
3. Section 1.5, Tool Availability Capability Check
4. Section 2, FAQ
5. Section 4, Visuals
6. Section 5, Featured Image Prompt
7. Section 6, Slug
8. Section 6.5, Excerpt
9. Section 7, Categories
10. Section 8, Tags
11. Section 9, SEO-Final Post Title
12. Section 10, Meta Description
13. Section 11, Focus Keywords
14. Section 12, JSON-LD
15. Section 13, Internal Link Discovery and Embedding
16. Section 14, External Link Recommendations
17. Section 15, Open Graph
18. Section 16, Optional Add-Ons, only if requested
19. Section 17, Pre-Flight Audit, thirty-one checks
20. Content Document body HTML, produced internally
21. Quick Paste Sheet, assembled from the values already produced
22. QUESTION BLOCK. Present every unknown. Stop. Wait for answers.
23. Write both files

Steps 1 through 21 run without interruption. Step 22 is the only pause. Step 23 happens once and produces final files.

Output Architecture

Two files, same turn, matching version stamps and matching slugs.

File 1: WOPPA_Support_[slug]_v2_3.md

Markdown. Contains the Quick Paste Sheet, the Paste Map, and Sections 1 through 17.

File 2: WOPPA_Content_[slug]_v2_3.html

Plain HTML. Not markdown, not a code fence. The Paste Map sits at the top as an HTML comment, which does not render and can stay in place after pasting or be deleted.

Why the Content Document is a plain HTML file

Earlier versions required the body to arrive in a single contiguous code fence, a rule written to stop a packager splitting a body across several fences. That rule was right about the outcome and wrong about the mechanism.

At roughly 264,000 characters the fence stops working. The reading surface falls back to plain text, and the single-copy workflow it was protecting dies. An HTML file is one contiguous block by definition, opens in any text editor, and select-all-copy yields exactly the body with nothing to strip.

The rule is the outcome: one file, one select-all, one paste.

Site Profile

WOPPA needs a small set of facts about the destination site. They change slowly and rediscovering them on every run wastes time and produces questions that did not need asking.

This profile ships configured for the author’s site. Any other publisher replaces these values before running, or supplies a site URL and lets WOPPA discover them.

Domain:            https://basilpuglisi.com
SEO plugin:        AIOSEO
Upload path:       /wp-content/uploads/YYYY/MM/
Categories:        AI Thought Leadership
                   Basil's Brand Blog
                   Building Blocks by AI (reserved for #AIgenerated output)
Author:            Basil C. Puglisi, MPA
Author URL:        https://basilpuglisi.com/
Job title:         Human-AI Collaboration Strategist and AI Governance Consultant
sameAs:            https://orcid.org/0009-0007-4747-152X
                   https://www.linkedin.com/in/basilpuglisi
                   https://medium.com/@basilpuglisi
                   https://x.com/basilpuglisi
Attribution:       #AIassisted

Internal page inventory. Framework and hub pages available as internal link targets.

HAIA ecosystem     /haia-the-human-artificial-intelligence-assistant-ecosystem/
HAIA-RECCLIN       /haia-recclin/
HAIA-CAIPR         /haia-caipr
Checkpoint-Based
  Governance       /cbg/
Factics            /factics/
HEQ                /heq/
AI disclosure      /content/

Verify each target before embedding. A profile entry is a candidate, not a guarantee, and a stale inventory produces a confidently wrong link, which is worse than a discovered one.

Three ways to run. Use the profile above. Replace it with your own. Or supply a site URL and let WOPPA discover the domain, plugin, categories, and page inventory, then confirm them in the Question Block.

Field Output Format

Every value the publisher pastes into WordPress uses the same shape. Two parts.

1. A numbered list, so the value can be read and reviewed
2. A fenced code block containing only the pasteable value

Nothing else goes inside the block. No character count, no bracketed note, no trailing comment, no label. Those belong in the prose around it. A character count copied into a meta description field is the exact failure this rule prevents.

Where a value is a single string, such as a slug, the numbered list is unnecessary and the block stands alone.

Applies to: slug, excerpt, title, meta description, focus keywords, tags, categories, featured image alt text, JSON-LD, and Open Graph.

Quick Paste Sheet

The first section of the Support Document. Every WordPress field, in the order a publisher fills them in the dashboard, each in its own code block, with nothing between blocks but the field name.

### Quick Paste SheetEverything you paste into WordPress, in dashboard order. Details and
reasoning for each value are in the numbered sections below.
Post title

[code block]

URL slug

[code block]

Excerpt

[code block]

Categories

[code block]

Tags

[code block]

Meta description

[code block]

Focus keyword

[code block]

Featured image

Alt text [code block]
Description [code block]

Visual 1: filename.png

Alt text [code block]
Description [code block]
(repeat per visual)

JSON-LD header script

[code block]

Open Graph tags

[code block]

Body HTML is the companion Content Document, a separate .html file.

Every value here is generated from the same source as its body section rather than retyped. Audit 17.32 verifies they match.

Publisher Paste Map

A table, fixed column order, identical in both documents except for the closing line.

The Support Document closes the map with: this document holds metadata. Do not paste it into the body editor.

The Content Document opens with the map as an HTML comment and closes it with: everything below this comment is the body. Do not paste this file into a metadata field.

Support Document Sections

Section 1. Confirm Content Received

One line acknowledging the content document is received and treated as canonical. Restate the post title exactly as supplied without paraphrase. State the approximate body word count.

State the title block as received, line by line.

Permitted-edit log. List every edit applied to the canonical body, by class, with a count. Where the conversion-artifact class was used, name each artifact type removed and how many instances.

If the content document is missing, that is not a question for the Question Block. Stop immediately and ask for it, because there is nothing to run.

Section 1.5. Tool Availability Capability Check

Declare, in a table: web search, web fetch, code execution, image rendering, file output. Available or not, and what each is used for.

Where a capability is missing, state the fallback. A missing capability produces a stated limitation in the audit, never a guess in the output.

Section 2. FAQ

Read the body. Produce questions a reader or an answer engine would ask, answered from the body’s own content.

Count. Four to eight. Under 1,500 words, four to five. Between 1,500 and 5,000, six to seven. Above 5,000, seven to eight. Where the post’s primary entity is a term, an acronym, or a framework name, one question is the definition query and it goes first.

Format. H3 question, paragraph answer. Forty to sixty words per answer. No first-person singular.

This exact text is reused in the Section 12 FAQPage schema and in the Content Document body. Three-way parity, verified at audit 17.6.

Accordion markup was considered and declined. Collapsed content has an unmeasured effect on extraction.

Section 4. Visuals

Every visual in the post gets three things: a filename, alt text, and a description. However many there are, one, two, or seven, each gets its own block.

Where the visual sits in the body. Mark the position with a visible placeholder paragraph, styled as a dashed box, carrying the filename:

<p style="font-size:16px;line-height:1.5;color:#8a1c1c;background:#fdf4f4;border:1px dashed #c9a0a0;border-radius:4px;padding:14px 16px;text-align:center;margin:26px 0 22px 0;font-family:Consolas,Monaco,monospace;">[ IMAGE: filename.png ]</p>

The placeholder must render. An HTML comment is invisible in Visual view, so a publisher sees only the caption and has no sign an image belongs there. A dashed box is visible in both Code and Visual view and cannot be missed.

Never write a fabricated src. The publisher deletes the box and inserts the image in its place.

What this section outputs, per visual:

Alt text stays short because the Description field carries the detail. A multi-part diagram does not need a 160-character alt text when its Description can run a paragraph.

Where the supplied content already carries visuals, carry them forward rather than proposing new ones. Where the body contains a mechanism, flow, comparison, or structure a diagram would clarify and none exists, produce one as SVG plus a PNG render.

Every visual’s alt text and description also appear in the Quick Paste Sheet, labeled by filename, because the media library fields are filled at upload time and the publisher should not hunt through seventeen sections for them.

Section 5. Featured Image Prompt

A paste-ready generative prompt, in a code block, that someone can drop into an image tool without editing.

Not a concept description. Not a brief for a designer. A prompt.

Include composition, palette, lighting, any visible text with its exact spelling, and aspect ratio. State the #AIassisted watermark spelling explicitly where a watermark is wanted, because generative tools drift on text.

Then alt text, eighty to one hundred twenty-five characters, and a long description.

Close with a verification step: check the generated image for spelling errors in any visible text before upload.

Section 6. Slug

Three to five meaningful words. Under sixty characters. Lowercase, hyphenated, no stop words. Primary entity front-loaded. Evergreen, meaning no dates or version numbers.

Field Output Format applies.

Section 6.5. Excerpt

Placed directly after the slug, because publishers fill this field third.

Format. One short lead line that states the hook. Then one paragraph, two to four sentences, that states what the piece argues and what the reader gets.

Field Output Format applies.

Section 7. Categories

Read the Site Profile category list first. Recommend from it, with a one-line rationale each, primary first.

Where the profile is absent, discover from the site. Where discovery is ambiguous or returns nothing, this becomes a Question Block entry. Never pick a plausible-sounding category.

Where a category exists for a different production class, say so and recommend against it if the content does not belong there.

Section 8. Tags

Eight to fifteen. Primary entity first, then the specific terms, then the broader ones.

Numbered list, then the comma-separated string in a code block.

Section 9. SEO-Final Post Title

Under sixty characters, counted and stated in the prose. Primary entity front-loaded. Benefit or intent in the verb.

Where the editorial title in the body differs from the SEO title, say so and state which goes in which field.

Field Output Format applies.

Section 10. Meta Description

One hundred fifty to one hundred sixty characters, counted and stated in the prose. Primary entity inside the first hundred characters. Framed as what the reader gets.

Field Output Format applies.

Section 11. Focus Keywords

Primary, then secondary.

Placement audit table. Each location reported as present or absent with evidence, never asserted: SEO title, slug, meta description, first body paragraph, at least one H2, featured image alt text, in-article image alt text, FAQ, JSON-LD, and body frequency across the whole piece.

Where a location is absent and that is correct, say why. A conceptual featured image should not force the keyword into its alt text.

Section 12. JSON-LD Header Script

Why schema is here. Three statements, and they are separate on purpose.

*Rich results on Google.* Schema is what produces an enhanced result. This is what schema was built for and it is uncontested.

*The entity graph, across every engine.* Consistent naming of people, organizations, and frameworks across the title, the body, the tags, and the JSON-LD, together with Person.sameAs pointing at durable identifiers, is what makes an author resolve as one entity rather than a recurring string. Any system that synthesizes an answer needs to know who it is citing. No engine has said otherwise.

*Google’s carve-out.* Google states that schema is not required to appear in its own generative features, and that those features run on the core Search index rather than a separate system. That statement is about Google. ChatGPT, Claude, and Perplexity have published no comparable position and none has said structured data is ignored. Do not generalize Google’s carve-out into a claim about every engine, and do not claim an AI citation effect Google has specifically disclaimed for itself.

The graph. Article, Person, Organization, FAQPage. A DefinedTerm node where the post’s primary entity is a term or framework name, with the common short form as alternateName.

creativeWorkStatus carries the AI attribution.

No placeholder tokens. The domain is supplied by the Site Profile, or it is a Question Block entry. No placeholder token is written into any output.

Reconciliation block. AIOSEO, Yoast, and RankMath all auto-generate Article and Organization schema, and RankMath now auto-populates FAQPage. State both resolutions: disable the plugin’s schema for this post, or remove the Article and Organization nodes and keep Person, DefinedTerm, and FAQPage.

Field Output Format applies.

Section 13. Internal Link Discovery and Embedding

Default behavior: links are embedded in the Content Document body at existing anchor text.

Read the Site Profile page inventory first, then discover for anything the profile does not cover. Verify every target before embedding, because a profile entry is a candidate.

Embed at first meaningful mention, at anchor text already present in the body. Never create new anchor text to hold a link.

Report a table: anchor phrase, target URL, body location.

Where a candidate exists but no natural anchor does, list it as available to the publisher and do not force it.

Where discovery returns nothing and no profile exists, that is a Question Block entry, not a silent omission.

The trigger modifier “without embedded internal links” reverts this section to a checklist and the Content Document carries no internal anchors.

Section 14. External Link Recommendations

Every external link in the body, verified at source, with the verification method stated.

Required attributes. Every external anchor carries target="_blank" and rel="noopener". Audit 17.26 checks this.

Report a table: anchor text, URL, verification status and date.

A URL that cannot be verified is not embedded. It becomes a Question Block entry.

Section 15. Open Graph Tags

Default output. og:title, og:description, og:type, og:url, og:image, og:site_name, og:locale, twitter:card, twitter:creator.

Title and description match Sections 9 and 10 word for word.

Field Output Format applies.

Section 16. Optional Add-Ons

Produced only when the trigger requests them. Reading time, pull quotes, schema validator pre-check.

Section 17. SEO / AIO Pre-Flight Audit

Thirty-one checks, each one numbered. An executor runs this list as written.

Every check reports evidence. A PASS with no evidence is the Audit-Faker pattern and fails the audit.

17.31 is load-bearing. An executor that writes its own audit will pass every check it wrote for itself, which is an audit of nothing. Renumbering, substituting, or inventing checks fails 17.31 regardless of how the other thirty resolve.

The Question Block

The only pause in the run. It sits after the audit and before any file is written.

What goes in it

Every value WOPPA needs and does not have. Every discovery that returned nothing or returned ambiguity. Every choice the publisher should make rather than the packager.

Typical entries: the domain when no Site Profile was supplied; the SEO plugin, which determines the reconciliation path; a category when discovery was ambiguous; the author schema when it is not in the profile; a focus keyword the publisher wants to set; an external URL that would not verify.

What does not go in it

A physical step the publisher takes after publication. Uploading an image is a step, marked in the body by a numbered comment, and no question is asked about it. Confirming the live post rendered is a step. These are post-publication notes in the audit.

Format

### QUESTION BLOCKThe run is complete. The audit is complete. These answers are needed
before the files are written.

[Question]
Why it is needed: [one line]
Where it lands: [section and field]

[Question]
...


Nothing has been written to file. Answer these and both documents
are produced in one pass.

After answers

Write both files. Do not re-run the audit unless an answer changed something the audit measured, and where it did, re-run and say so.

Where there are no questions, say so in one line and write the files.

Content Document Production

Executes last, after every section it depends on.

Step 1. Capture the title block

Take the supplied title block word for word.

Step 2. Convert the title block

H2 rather than H1, because the WordPress theme renders the post title field as the page H1.

Where a source title block has more or fewer elements, map by position and structural role.

Step 3. Convert the body headings

Perform the demotion in reverse order. H5 to H6 first, H1 to H2 last.

Forward order collides. Converting H1 to H2 first means the next pass converting H2 to H3 also catches every heading the first pass just created, and each subsequent pass compounds it until every heading in the document sits at one level. Every character is present, nothing errors, and the document is wrong in a way that only shows up in a page count or a heading audit.

Audit 17.25 catches it after the fact. Reverse order prevents it.

Step 4. Convert the body

Paragraphs to <p> with inline styles. Lists to <ul> and <ol>. Emphasis to <strong> and <em>. Tables wrapped in <div style="overflow-x: auto;"> with inline-styled <table> inside.

Inline CSS quoting. Inner string values use single quotes: style="font-family: Georgia, 'Times New Roman', serif;". A double-quoted font name inside a double-quoted style attribute breaks the attribute boundary and renders as visible text. Audit 17.30 checks this.

Inline styles rather than classes, because a theme can strip or override a class and inline styles survive the paste.

Step 5. Handle images

Every source image becomes a visible dashed-box placeholder naming its file, per Section 4. No fabricated src, ever.

Diagrams WOPPA produced are delivered as files alongside the documents and referenced the same way.

Step 6. Insert links

Internal at existing anchor text per Section 13. External at existing anchor text per Section 14, with target blank and rel noopener.

Never create new anchor text to hold a link.

Step 7. Append the FAQ and the attribution

FAQ block after the body, before the attribution. H3 questions, paragraph answers, matching Section 2 word for word.

Then #AIassisted as the final line.

Step 8. Write the file

Plain .html. The Paste Map as a leading HTML comment carrying the procedure, the figure upload steps, and the featured image step.

<!--
WOPPA CONTENT DOCUMENT  |  [slug]  |  v2.3
[Post title]
WHERE THIS GOES IN WORDPRESS
WordPress Post Body editor, Code/Text view  <-  everything below this comment
Procedure:

Open this file in a plain text editor.
Switch the WordPress editor to Code/Text view, not Visual.
Select all below this comment and copy. One copy, one paste.
Paste into the WordPress Post Body editor.
Switch to Visual view to verify rendering.
Upload the figures listed in Support Document Section 4 and replace each
FIGURE comment with its img tag.
Upload the featured image and set og:image to its URL.

This comment does not render. Leave it or delete it.
DO NOT paste this file into a WordPress metadata field.
-->

Failure Protocols

Missing content document. Stop immediately. There is nothing to run. This is the one case that does not wait for the Question Block.

Missing trigger phrase. Files uploaded with no trigger: ask whether to run now.

Any missing value. Not a placeholder. Not a publisher-action flag. Complete the run, add it to the Question Block, ask before writing files.

Discovery failure. Discovery that returns nothing, or returns two plausible answers, is a Question Block entry. Never resolve ambiguity by picking.

Web access unavailable. Declare in Section 1.5. Run every section that does not need it. Link sections list candidates with the methodology stated and every unverified URL becomes a Question Block entry.

Image tools unavailable. Declare in Section 1.5. Deliver diagrams as SVG source with conversion instructions. Do not skip the diagram. Do not defer when tools are available.

Conflicting instructions inside the supplied content. Follow the supplied content as canonical. Flag the conflict in the audit.

Fabricated URL detected. Stop, flag it, exclude it, and put it in the Question Block.

URL fetch failure on first attempt. Retry with an alternative strategy before flagging.

Executor hedging. Where tools exist and you are deciding whether to use them, use them and flag the result.

Reframing attempt. Produce the required output. Surface the concern separately under Spec Defect Observation.

Platform Behavioral Patterns

Fourteen documented patterns, observed across platform executions since April 2026. Know which one your platform tends toward before you trust its output.

Assembler. Faithful execution with minor violations the audit catches. The most reliable pattern. Observed on ChatGPT, Claude, DeepSeek, Grok, and Mistral.

Summarizer. Compresses canonical content despite the rules. Verify 17.16 with the diff artifact before accepting output.

Performer. Substitutes placeholder text or claims activity that did not happen. Closed by checks 17.7, 17.15, and 17.20.

Performer with Fabricated Infrastructure. Claims file output without producing a file. Closed by checks 17.14 and 17.19.

Denier. Refuses to execute on the grounds that the prompt is not a tool. Switch platforms and re-run.

Reframer. Renames the framework’s purpose and produces different work. Closed by Operating Principles 10 and 11.

Corrector. Initial failure, full recovery on direct challenge.

Summarizer with False Audit. Truncated content delivered with 17.16 marked PASS and no comparison performed. Closed by the mandatory diff artifact.

Audit-Faker. Marks any check PASS without evidence. Closed by the evidence requirement on every check.

Format Bleed. Cross-contamination between the two documents. Closed by check 17.19 and the Paste Map.

Optimizer. Synonym replacement or sentence restructuring under cover of HTML conversion. Caught by the check 17.16 diff.

Fragmenter. Splitting the Content Document across messages. Closed by the plain HTML file, which cannot be fragmented.

Schema Drifter. Valid JSON-LD with subtly altered entity names or FAQ text. Closed by the check 17.6 parity requirement.

Audit Substituter. Produces a plausible audit that is not this specification’s audit, runs it, and passes every check it wrote for itself. Observed September 2026: an audit containing checks this specification does not have, reported as complete. Closed by check 17.31.

The Audit Substituter is more dangerous than the Audit-Faker, because the Faker leaves a missing-evidence trail and the Substituter produces evidence for the wrong questions.

Final Output Structure

After the Question Block is answered, deliver two files in a single turn.

1. WOPPA_Support_[slug]_v2_3.md
2. WOPPA_Content_[slug]_v2_3.html

Plus any diagram files produced in Section 4.

Close with:

“HAIA-WOPPA v2.3 complete. Support Document for metadata and governance. Content Document as a plain HTML file, one select-all, one paste. [N] visuals require upload before publishing, marked in the body by numbered comments.”

Basil C. Puglisi, MPA

A Human-AI Collaboration

#AIassisted using the HAIA Ecosystem

Frequently Asked Questions

What does WOPPA stand for?

WOPPA stands for WordPress Optimization Publication Prompt. WordPress is the destination, so the output is shaped for one platform. Optimization means the metadata that makes a post findable, not the writing. Publication means it runs at the moment of publishing. Prompt means it is instructions rather than software.

What does HAIA-WOPPA actually produce?

Two documents. A Support Document carrying every WordPress metadata field in its own copy block, including title, slug, excerpt, categories, tags, meta description, focus keyword, alt text, schema, and Open Graph. And a Content Document carrying the body as a plain HTML file that takes one select-all and one paste.

Does WOPPA change my article?

No. The article is treated as canonical and the only permitted edits are mechanical: converting headings, wrapping tables, inserting links at existing anchor text, and replacing source images with numbered placeholders. WOPPA does not evaluate the content, improve it, or rewrite a sentence of it.

Why does it stop and ask questions?

Because a missing value is a question rather than a placeholder. The run completes every section and the full audit internally, then pauses once before writing any file to ask what it needs: the domain, the SEO plugin, an ambiguous category, an upload path. Nothing ships carrying a token to swap later.

What platform does it run on?

Any AI platform that accepts a file upload. Claude, ChatGPT, Gemini, Perplexity, Grok, DeepSeek, Kimi, MiniMax, and Meta AI have all been used. It is a prompt rather than software, so there is nothing to install and no account beyond the one you already have.

Is it free to use?

Yes, under Creative Commons, with one condition. The exact hashtag #AIassisted stays at the end of every body it produces. Not a variant, not a substitute, and not removed. That tag is how transparent AI-assisted publishing stays visible on the open web.

Do I need HAIA-CORE to use WOPPA?

No. WOPPA takes a finished article from any source, written by hand, drafted with AI and edited, or converted from a Word document. It does not care where the content came from or whether it is any good. Evaluating content is a different question answered by a different tool.

#AIassisted