Troubleshooting & FAQ

Applies to Antradus AI Lite, Publisher & Studio v2.7.0 · Last updated 2026-08-26

Setup & license

A feature shows a 🔒 lock — why?
A lock means the feature belongs to a plan above the one you’re on, and there are two possibilities. On Lite (or with a license that isn’t active), the locked rows are the Publisher ones: bulk publishing, YouTube to Article, products from photos, albums, web search, merge, internal linking, SEO clusters, AI Search and auto-fill. On Publisher, the only locked things are the two Studio tabs on the Bulk Publishing screen — Podcast to Articles and Episode Assets. Either way, click the lock to see what it costs.

I can’t find the panel while editing a post.
Classic editor: it’s below the content area. Block editor: click the anchor ⚓ icon in the top toolbar (or three-dots menu → Antradus AI) — it opens in the right sidebar.

"An older copy of Antradus AI Lite was automatically deactivated…"
Expected: a pre-2.0 standalone Lite was installed alongside the new plugin, which detected and deactivated it to prevent conflicts. The new plugin includes everything it had; your settings carried over.

Generation problems

"The AI model took too long."
The provider timed out — usually a slow model or provider congestion. Try again; if it persists, pick a faster model or generate without web search.

"The AI hit its output limit before finishing."
The model spent its budget thinking instead of writing. Reasoning models — gpt-5.x, o-series, Gemini thinking models, deepseek-reasoner — split one budget between their internal reasoning and the words you get, and a long or complicated request can leave nothing for the answer. The plugin already asks for far more room than the writing needs, so if you see this the request itself is the problem: ask for fewer articles, select less text to rewrite, or switch to a model that reasons less (gpt-4o, claude-haiku, gemini-flash) in the Antradus AI menu. Nothing part-written is ever published.

Before v2.5.1 this appeared as "Empty response from the AI provider" or "the AI reply could not be understood — try a stronger model", both of which pointed at the wrong thing. Switching to a stronger model made it worse, because stronger usually means it thinks more.

The AI refused and asked me for a source link.
Your topic is unknown to the model (an obscure local business, a person with no public record). It refuses rather than inventing facts. Paste a source URL into the URL field and run again.

"This website is blocking outside access to its content."
The source site blocks automated readers (common: academic journals, paywalled news). Try an open-access copy — Google Scholar, the author’s own site, or a preprint server.

Fetching models fails on the settings page.
Almost always the API key (typo, revoked, or no credit on the provider account). Re-paste the key and check your provider’s billing page.

Web search did nothing / errored.
Check the capability matrix: DeepSeek has no web search; OpenAI needs a search-capable model; some OpenRouter models ignore search results.

Images

Images fail or silently save as PNG instead of WebP.
Image generation needs an image-capable provider (OpenAI / Gemini / OpenRouter — not Claude or DeepSeek) and an Image Model selected. WebP conversion additionally needs the PHP GD extension — ask your host if PNGs appear unexpectedly.

Album generated 7 of 9 images.
Individual failures don’t stop the album. Regenerate later or accept 7 — the succeeded images are all in your Media Library.

SEO

Yoast/Rank Math fields look empty after generating.
The values are stored server-side even when the on-screen fields lag. Save the post and refresh — they’ll be there. (Both plugins must be active to see their panels, obviously.)

The SEO button says "💾 Save as draft first".
The post has never been saved, so there’s nothing to attach meta to. Save as draft, then generate.

Why is the focus keyword different from my topic text?
By design — the AI derives the ideal keyphrase from the content instead of echoing your raw input. It converges on the same search intent with cleaner phrasing.

Bulk publishing

Jobs sit in "queued" and nothing happens.
Almost always this: WordPress has no background service. Its scheduler only runs when someone loads a page on your site, so on a local install, a staging site, or a new site with no visitors, a closed tab means no progress at all. The ⏳ Queue tab says so directly — the box at the top reports when the background runner last ran and warns when work is waiting with nothing processing it.

Fix it now: open the Queue tab, press ▶ Process now, and leave it open — that route is the fastest anyway (a couple of articles a minute in the background, faster with the tab open). For unattended overnight batches, ask your host to run a real cron job every minute (wp cron event run --due-now, or a scheduled request to the wp-cron.php URL the Queue tab shows you). See Bulk Publishing → what happens when you close the tab.

A batch has failures.
Open ⏳ Queue — each failed row shows its reason (usually a provider timeout). Retry failed re-queues just those.

A post says done but has a warning about its image.
The article succeeded; only its image failed. Add one manually or regenerate from the post editor.

My CSV rows were skipped.
Duplicate protection: a post with the same title already exists. That’s a feature — re-running a CSV never doubles content.

Scheduled (drip) posts didn’t publish on time.
Same WP-Cron caveat as above — publishing happens on visits. Any regular traffic (even your own) is enough; for a brand-new empty site, visit it or set up a ping service.

Costs & privacy

Where does my money actually go?
To your AI provider, at their rates, on your key. The plugin adds no fees and no middleman. The Reporting page estimates costs per post; your provider’s dashboard is the invoice of record.

Does the plugin send my content anywhere?
Only to the AI provider you configured, only when you click a generate button. No other servers see your keywords, articles, or keys.

Do free OpenRouter models train on my prompts?
Often, yes — providers serving free models commonly retain and/or train on inputs. Use a paid model for sensitive content.

Still stuck?

Contact support via Antradus AI → Contact Us in your admin (it includes your environment details automatically), or through your account page.