Coverage — Answer-Engine Audit

The Coverage view answers one question: across all the articles your AI answer files describe, what has actually been run against each one?

It has three modes. GEO is the answer-engine audit described below. SEO turns the same table into a bulk pass over your SEO titles, meta descriptions, focus keywords and the alt text on your pictures — see Filling in SEO fields in bulk and Describing your pictures. Comments lists every page and post that has review comments and lets you delete them in bulk — see Deleting review comments in bulk.

In GEO mode it puts every one of those articles in a single table — whether the GEO analysis has ever run, how the article scored, how many questions and answers it has, whether anyone has reviewed those answers, and whether its meta description, SEO title, focus keyword and pillar mark are filled in. From the same table you can run the GEO analysis over a batch of articles at once, or mark existing stored answers as reviewed in a batch without re-running AI.

Where to find it

Open the Content Pipelines tab and click the Coverage pill in the row of view links at the top. The Mode pills just under the heading switch between GEO, SEO and Comments.

Coverage is for SyteOps administrators only. It is deliberately not part of the Review Portal — a reviewer working on one article already has an Analyze GEO button in that article's GEO panel, and that is a different, one-article-at-a-time action. Coverage is the whole-library view.

Which articles are listed

Coverage lists exactly the content your answer engines describe — the same content types you picked under Content Pipelines → Answer Engine → Answer-engine publishing → Content types. Change that setting and Coverage follows it, so the two can never disagree about what is in scope. (See Choosing which content types are covered.)

Beyond that:

• Published, scheduled, draft and pending articles are all listed and included in bulk runs by default, so work started on a draft is not repeated in the Review Portal. Switch Include drafts & scheduled off to limit the list and every bulk run to published articles (Clear filters switches it back on).
• Password-protected articles are never listed. They are not public, so nothing SyteOps publishes describes them.

What each column tells you

Column	What it shows
Article	The article title, linked straight to its editor so you can open it without hunting for it.
GEO analysis	Which of the four states the analysis is in — see The four GEO states below.
Readiness	The readiness score from the last analysis: how well the article is structured to be picked up and cited by AI answer engines.
Answers	How many of the article's recorded questions have an answer written for them. A question with no answer is not much use to an answer engine.
Reviewed	Whether someone has ticked I have reviewed these answers for this article. Until they have, its answers are held back from every answer-engine surface.
Awaiting review	How many of the article's answers are currently being held back, waiting for that confirmation.
Meta description	Whether the article has one. This is taken from the article's excerpt.
SEO title	Whether a search-result headline has been set.
Focus keyword	Whether a focus keyword has been set.
Pillar	Whether the article is marked as a pillar (cornerstone) article — see Pillar articles.

The SEO title and Focus keyword columns appear only when you have an SEO tool integration switched on. With no SEO plugin connected there is nowhere for those values to live, so the columns stay out of the way.

The four GEO states

State	What it means
Never analyzed	The GEO analysis has never run on this article. Nothing has been recorded for it at all.
Stale	The analysis ran, but the article has been edited since. What was recorded describes an older version of the text. With Analyze GEO automatically on, a draft or scheduled article is re-analyzed by itself a few minutes after its text changes; a published article stays Stale until someone analyzes it here or in the Review Portal.
Current	The analysis matches the article's current words. Nothing is owed here.
Analysis failed	The last attempt errored. The row shows the reason and a Retry link so you can try that article again.

:::note "Never analyzed" is not the same as a score of zero A readiness score of 0 is a result — the analysis ran and found the article poorly structured for answer engines. Never analyzed means no analysis has happened, so there is no result to read. Treat the first as something to fix in the article, and the second as something to run. :::

Filtering the list

Three sets of filters sit above the table, and they combine:

• By GEO state — Never analyzed, Stale, Current or Analysis failed. Filtering to Never analyzed is the usual way to find everything that has never been looked at.
• By gap — Answers awaiting review, No meta description, No focus keyword or No SEO title. Each one narrows the table to articles missing that single thing, so you can clear one kind of gap across the whole library in one sitting. The keyword and SEO title filters appear only when an SEO tool integration is switched on.
• Include drafts & scheduled — on by default. Turn it off to narrow the list, and the bulk runs, to published articles only.

Clear filters puts you back to the full list.

Analyzing several articles at once

Tick the articles you want — there is a Select all on this page box if you want the visible page — and click Analyze N selected.

Before anything runs, a confirmation dialog tells you exactly what you are about to spend:

• The AI provider that will be called.
• The model it will use.
• The maximum tokens per article.

It also states plainly that this spends money on your own AI account, one call per article. SyteOps does not estimate a dollar figure, because token pricing differs per provider and changes without notice — a number there would look authoritative and be wrong. The provider, model and ceiling are the facts you need to judge the cost yourself.

:::warning The "waiting for review" count will go up A run writes new questions and answers, and new answers are held back until a person reviews them. So the number of articles with answers waiting for review will increase after a bulk run — often sharply, if you analyzed a large batch.

That is the review gate working exactly as designed, not a fault. Work through those articles in the Review Portal and tick I have reviewed these answers on each; the answers reach your answer-engine surfaces at that point. See Reviewing the answers before they go live. :::

Accepting stored answers as reviewed

Use Accept answers as reviewed when articles already have stored question-and-answer pairs and you only need to confirm them for publication. It uses the same row selection and Select all on this page control as bulk analysis.

Before the run starts, a confirmation dialog explains that this publishes answers that were held back for review onto your answer-engine surfaces. It does not run AI and does not spend against your AI account.

The run works the same way as analysis — one article at a time, a Stop control, and at most one bulk Coverage run at a time across the site. Articles with no stored answers are skipped, as are articles somebody currently has open in the Review Portal.

When the run finishes you get a summary of what was accepted, what failed and what was skipped. Reload the page to see the updated Reviewed column. Until answers are reviewed this way (or in the Review Portal), they stay off public answer-engine surfaces.

What a bulk run will and will not do

Both Analyze and Accept share the same run mechanics — one article at a time, a Stop control on the button you started, and at most one bulk Coverage run at a time across the site. If someone else is already running one, you are told to wait.

The limits below apply to both run types unless a bullet says otherwise:

• At most 25 articles per run. Ask for more and the run is refused outright — it is not quietly trimmed to the first 25.
• Articles are processed one at a time, so the progress you see is real progress.
• Clicking Stop on the active button halts the run after the article in progress finishes.
• Only one bulk run can be in flight at a time on the site.
• Some articles are skipped, and every skip is reported with its reason:
  • An article somebody currently has open in the Review Portal is skipped. Its checkbox is disabled, and a line under the buttons says how many rows on the page cannot be selected and why.
  • Analyze only: an article whose exact current text already failed is skipped rather than being sent to the provider to fail identically and bill you again. Editing the article makes it eligible again. The Retry link on a failed row clears that block for a single article.
  • Accept only: an article with no stored answers is skipped.

After an Analyze run

When an analyze run finishes you get a summary of what was analyzed, what failed and what was skipped. Skipped articles are named (up to five, then a count) with the reason, and any skip turns the notice into a warning. Reload the page to see the updated table. You may lose a paid completion you have already started if you stop mid-run, but you never have to wait out a batch you changed your mind about.

After an Accept run

When an accept run finishes you get a summary of what was accepted, what failed and what was skipped. Reload the page to see the updated Reviewed column. Accept does not spend against your AI account.

Filling in SEO fields in bulk

Switch Mode to SEO and the table changes to four columns — SEO title, Meta description, Focus keyword and Structure — showing what each article holds today, with a Missing badge where there is nothing. The row picker, Select all on this page, the gap filters and the Include drafts & scheduled toggle all work exactly as they do in GEO mode. The Structure column is a read-only report on the article body itself — nothing here proposes or writes anything — giving a one-line summary of its headings, images and links (hover it for the full list of anything worth a second look), with four filter pills that narrow the table to a specific problem: Multiple H1 (more than one top-level heading), Heading order (a heading level skipped, such as jumping from an H2 straight to an H4), Images without alt (a picture in the body with no alt text; a picture deliberately marked decorative, with an empty alt plus role="presentation", role="none" or aria-hidden="true", is not counted), and No internal links (nothing in the body linking to another page on your own site); because it reads the article's content directly, it works even with no SEO tool integration switched on. Images without alt reads the article and skips a picture marked decorative, while the Pictures needing alt text pill reads the Media Library and still counts a decorative picture whose library entry is blank; a decorative marker on a wrapping figure or link is not recognized, so that picture still counts.

This is a two-step workflow, and the split is deliberate: unlike answers, an SEO title appears in a search result the moment it is stored, so nothing is written until you say so.

Step 1 — Propose

Tick the articles you want and click Propose for N selected. A confirmation dialog names the AI provider, the model and the maximum tokens per article, states that this spends money on your own AI account, and says plainly that nothing is written yet.

The run works one article at a time, with a Stop control, and produces a suggestion for each field that is empty:

• Focus keyword — one short phrase drawn from the article body.
• SEO title — a search-result headline, kept under 60 characters, and written around the focus keyword when the article has one (or the one just suggested for it).
• Meta description — the same writer the Review Portal's Generate with AI button uses.

Two optional tick boxes sit beside the buttons and change what the run is allowed to touch:

• Re-propose existing keyword — also suggests a keyword for articles that already have one.
• Overwrite existing values — also suggests for fields that are already filled, in all three columns.

Both are off by default, so a plain run only fills gaps and never proposes over work somebody already did. Whichever you tick is named in the confirmation dialog before the run starts.

Articles are skipped, with the reason reported, when somebody currently has them open in the Review Portal, when they are out of scope, or when every field the run may touch is already filled.

Step 2 — Review and accept

Each suggestion appears under the current value in its own cell, with a tick box. Everything a run proposed starts ticked.

• Untick anything you do not want — that is Skip for that field.
• Accept all fields and Skip all fields tick or untick everything on the page at once.
• Click Accept N selected to write the ticked fields.

The accept run does not run AI and does not spend against your AI account. Its confirmation dialog tells you how many fields will be written and reminds you that they appear in search results as soon as they are stored.

Three things are refused rather than written, and each is reported:

• The value changed since it was proposed. If somebody edited that SEO title in the Review Portal after the suggestion was made, the suggestion is discarded rather than written over their work. The article goes back into the propose queue.
• The field is already filled and this run was not started with Overwrite existing values ticked.
• No SEO tool is switched on, so there is nowhere for the value to be stored. Turn one on under Integrations first. (Proposing still works without one — only writing needs it.)

Suggestions are kept with the article, so you can propose today, reload, and accept tomorrow.

What the SEO mode writes

Accepted values go through exactly the same path the Review Portal uses, so an article filled in here and one filled in by a reviewer end up in the same state. The meta description is stored as the article's excerpt and mirrored into your SEO plugin; the SEO title and focus keyword go into whichever SEO plugin you have connected.

Both SEO runs share the same limits as the GEO ones: at most 25 articles per run, one at a time, one bulk Coverage run at a time across the whole site, and a Stop control that halts after the article in progress. The picture runs share all of that and add a second ceiling of their own: at most 50 pictures per run.

Describing your pictures

The same SEO table has two more columns — Pictures and File name — and its own pair of buttons below the SEO ones.

Pictures lists every image the article shows that also lives in your media library: its featured image and every picture in the body. A badge says how many still need work, meaning the picture has no alt text, has alt text that is really just the file's name, or has no title on its media record. If some of the article's images are hotlinked from another site, a note says so — there is no media record for those, so nothing SyteOps writes can reach them.

Like the SEO fields, this is a two-step workflow.

Step 1 — Describe

Tick the articles you want and click Describe pictures. The confirmation dialog tells you how many pictures will be described, because that is what you are billed for — one AI call per picture, not per article. It also names the provider, the model and the token ceiling, and says plainly that nothing is written yet.

The AI looks at the picture itself and, using the article around it for context, suggests three things: alt text (one plain sentence for someone who cannot see the image), a caption, and a title for the media record.

• At most 50 pictures per run. Ask for more and the run is refused outright — it is not quietly trimmed.
• A picture that is already described is skipped, and so is one that already has a suggestion waiting for you.
• A picture the AI could not read, or whose file type it does not accept, is remembered so the next run does not pay to be told the same thing again. Replacing the file makes it eligible once more.

One optional tick box changes what a run is allowed to touch:

• Overwrite existing picture text — also describes pictures that already have alt text or a title.

Step 2 — Review and accept

Each suggestion appears in the Pictures cell under the picture's file name, with a tick box, already ticked. Accept all picture fields and Skip all picture fields tick or untick everything on the page. Click Accept picture text to write what is left ticked. That step runs no AI and spends nothing.

:::warning Alt text belongs to the picture, not to the article If the same picture appears on three articles, changing its alt text changes how all three describe it. SyteOps will not replace alt text on a picture another article is showing unless you tick Overwrite shared pictures — and the confirmation dialog says so before the run starts. :::

Suggestions are refused rather than written, with the reason reported, when:

• The value changed since it was suggested — somebody typed real alt text in between, and their wording wins. That picture goes back into the describe queue.
• The field is already filled and the run was not started with Overwrite existing picture text ticked. The exception is text that is only the file's own name, which is replaced without asking, because nobody chose it.
• Another article shows that picture and Overwrite shared pictures was not ticked.

The File name column

This column reports pictures whose media record says nothing the file name does not already say — an IMG_4821.jpg with no title, for example. It is report only. Renaming a file means finding and updating every place it is referenced, so it is not done from this screen; each name links to that picture in the Media Library.

Deleting review comments in bulk

The Comments mode is a clean-up tool for the comments people leave in the Review Portal — not WordPress comments. It lists every page and post that currently has at least one review comment, whatever its content type and status — published, scheduled, draft, pending, private or a custom status. It is not limited to the content types covered by the answer engines, or to the types the Review Portal currently opens, so comments left on content from a source you have since changed still show up. Trashed items, revisions and media files are not listed.

The most recently commented items come first. For each one the table shows its type and status, how many comments it has in total, how many of those are replies, how many are still open (top-level comments nobody has marked resolved), and the date and author of the latest comment. The title opens the item in the Review Portal.

To delete comments:

1. Tick the pages and posts you want, or click Select all on this page.
2. Click Delete comments on selected.
3. Read the confirmation. It tells you how many pages and posts and how many comments in total will be removed. It removes everyone's comments, including replies, and it cannot be undone.
4. Confirm. The deleted rows leave the table. If any page or post could not be cleared, a message names it and the rest are still deleted.

Each page or post that is cleared gets one entry in its Review Portal activity history recording who deleted the comments and how many. Only SyteOps administrators can use this mode. That is stricter than the Review Portal's own Delete all comments button on a single article, which any WordPress administrator can also use.

The confirmation shows the counts from when the page loaded. If someone adds a comment in the meantime it is removed too, and the result tells you how many more were removed than shown. If the connection drops part-way, the message says some of the remaining pages may already have been cleared — reload to check.

Where the run is recorded

Every bulk run writes one row in the Content Pipelines Log — one row for the run, not one per article, so a 25-article batch does not bury the rest of your history.

Filter the runs list with the Coverage source pill to see only these runs. Each row records how many articles were analyzed, accepted, proposed for, described or written, how many failed and how many were skipped, and — for SEO and picture runs — which fields moved on each article. Runs that call AI also record the model that was billed, so AI spending is answerable after the fact, not just at the moment you approved it.