BlogCraft troubleshooting.
Why Generate is greyed out, why your outline reset, why 2,310 words counts as 2,500, and why your WordPress post arrived without its meta description.
Eight things that look like bugs and aren't. Each one has a cause, and each one has a fix.
Generate is greyed out
The big orange Regenerate full article button is disabled, and underneath it says “✓ Up to date — edit the outline or a setting to regenerate”.
That's the point. Once an article has been generated, the button stays locked until something about the recipe actually changes. Nothing has changed since the last run, so regenerating would burn credits to produce the same article again.
To unlock it, change any one of these:
- The outline — edit an H2, add or remove a section, reorder them, change a schema hint.
- The length — a different preset, or a custom word count.
- Any writer setting — tone, GEO, E-E-A-T, entities, links, CTA, FAQ, meta.
The moment you do, the status chip beside the outline flips from a lime Generated · up to date to an orange Edited · regenerate to apply, and the button lights up.
If you just want to improve what's already there, don't regenerate — you'd throw away the whole article and pay full price for a new one. Select the passage that bothers you and use Re-tune for 2 credits.
My outline vanished
You'd hand-edited your H2s, went back to Step 1 to change the keyword, and came back to find a completely different outline.
The Step 2 research pipeline is keyed to your keyword. Change the keyword and confirm a title again, and ScoutRival re-runs keyword expansion, re-fetches the search results, and builds a brand-new outline from them. Your edited sections are replaced — they were built for a different keyword.
The same applies to your title variants: asking for suggestions again gives you five fresh ones and replaces the old set.
- Settle the keyword first. It's the seed for everything downstream — the research, the outline, the internal keyword lists.
- Then edit the outline. Outline edits are free and never reset themselves; only a keyword change re-runs the research.
- Intent (informational, commercial, transactional) shapes the titles you're offered and the outline that gets built. Set it before you confirm a title, not after.
I asked for 2,500 words and got 2,310
That's in band, and the article is fine.
Language models cannot count their own words. Every serious AI writing tool works to a range rather than an exact integer, and so do we: the band is ±5% of your target, with a minimum of ±150 words. For a 2,500-word ask, anything from 2,350 to 2,650 is on target — and 2,310 is a hair under, which is why the editor tells you so honestly rather than pretending.
When a draft does come in under the floor, ScoutRival has already tried to expand it, and the note explains what it did: “Came in just under target — we auto-expanded the draft as far as it could go without padding.” It stops short rather than filling the gap with waffle, which would cost you Content Score points and reader patience both.
If you genuinely need the length, use Lengthen on a couple of thin sections — it deepens existing points rather than padding. More on choosing a length →
Generation seems stuck
The five phase rows have been sitting on FAQ + schema or Entity coverage for what feels like a long time.
Outline locked, Writing and AI-visibility pass tick over on an estimate, not on real progress — there's no live stream to read from, by design. The last two rows are the honest ones: FAQ + schema and Entity coverage stay on “running” until the article genuinely exists. So a long pause on the final rows is the pipeline actually working, not a hang. When all five turn green, the article is really there.
A full run takes roughly two to three minutes. Things worth knowing:
- Don't refresh the page. The article is created on the server whether or not your tab is watching, but refreshing means you lose the redirect and have to find it in your library.
- Longer articles take longer. A 6,000-word piece is not a 1,500-word piece with a bigger number.
- If it fails outright, you'll be told. A silent forever-spinner is not one of the failure modes — errors surface as an error.
My entity count looks wrong
The Content Score's Entity coverage row shows a different number from what you expected, or it changed on its own a second after the page loaded.
- It loads a moment after the editor. The Entities card fetches its list on open and then feeds that count into the score. If you watched the score tick, that's the two numbers agreeing — deliberately, so you never see one figure in the card and a different one in the score.
- It counts unique entities, not mentions. Naming the same tool eleven times is one entity.
- Your own products don't count. The extractor deliberately skips references to your own brand's products — they do nothing for entity-density SEO. Neither do generic words like “customer” or “company”.
- Names get canonicalised. “The Bedford template” and “the Bedford” collapse into one entity.
- It's capped at 40. Beyond that it stops counting — which is well past the 20 you need for full marks.
If it looks stale after a big edit, hit Re-scan entities on the card. It's free.
Internal links found nothing
The Internal links half of the Linking strategy card says “No matching prior articles found.”
This is the expected outcome for most people, and it isn't a failure. The card searches your Content library — drafts saved out of Compose — and your BlogCraft articles don't go into it automatically. If you've never saved anything there, there is nothing for it to match against, and the card will be empty every time.
External citations, in the same card, work normally and are the half that actually feeds your Content Score. The full explanation →
My WordPress post lost its meta description
It didn't lose it. It never had it.
Publishing sends the title and the article HTML. Nothing else. No meta title, no meta description, no JSON-LD, no categories, no tags, no slug, no excerpt, no featured image.
Everything ScoutRival generated around your article lives in the SEO pack under the editor, and you carry it across yourself: Copy meta on the SERP preview card, Copy code on the FAQ schema card. Or export the article as Markdown, which appends the whole SEO pack to the file. What gets sent, in full →
My share link still works after I revoked it
Wait five minutes, then check again.
The public share page is cached at the edge for five minutes so it loads quickly for whoever you sent it to. Revoking kills the token instantly on our side, but a cached copy can keep being served until that cache expires. The app's confirmation dialog says “immediately”; the truthful answer is within five minutes.
There is no way to force the cache to drop early. If a link needs to be dead right now, revoking it is still the right move — just don't rely on the first refresh to prove it worked. More on share links →