AI Assisted Article Generator - Curating sources and search terms for scheduled articles using Topic Hoppers

A Hopper is “a curated list of source URLs on a topic, plus a scheduled brief from them”. This guide walks through setting one up end-to-end, from creating the Hopper to seeing the first article published unattended.

Hopper is a Premium-tier feature. Trial and Standard tiers see the menu hidden. This guide assumes you're on Premium or Enterprise.

The Hopper concept

A Hopper has three moving parts:

What the Hopper produces is briefs — one brief per cycle. The article-gen pipeline then runs on each brief the same way it runs on a one-off. Optional Auto-generate and Auto-publish settings make the whole loop unattended.

Step 1 — Create a Hopper

1. Go to Components > Article Generator > Hoppers and click New.
2. Topic tab — fill in:
   • Name — the editorial label, e.g. “UK politics — Andy Burnham”.
   • Description — a short topic frame the AI uses when synthesising the brief title. Keep it crisp.
   • Keywords — comma-separated. Inherited by every brief this Hopper emits.
   • Always-include source URLs — URLs that ride on every emitted brief alongside whatever the watch finds. Leave empty for most cases.
   • Authoritative sources for this topic — per-Hopper authoritative-domain hints (e.g. bbc.co.uk, nature.com). Sub-domains match. Used to prefer publisher-grade results in the candidate cluster step.
3. Editorial Defaults tab — tone, reading level, target word count, content structure, citation mode, banner icon, banner theme. These ride on every brief the Hopper emits, so you set them once.
4. Production tab — the watch + cadence:
   • What to watch — Search query / RSS feed / URL set.
   • Watch value — the actual query / feed URL / URL list.
     • Search query: the words you'd type into Google. Hopper expands this internally with “latest news” / “this week” time-bias variants, plus a Serper news-endpoint variant if Serper is your provider.
     • RSS feed: a single http(s):// URL. RSS 2.0 and Atom 1.0 both supported.
     • URL set: one URL per line, up to 20. Each URL becomes one candidate.
   • Cadence — Manual, Daily, 3x weekly, 2x weekly, or Weekly. Manual fires only from the Run Now button on the Candidates tab; the rest are driven by the scheduled task.
   • Time of day — shown when cadence is non-manual. Site-timezone HH:MM at which the cadence fires. Daily fires AT that time each day; weekly variants use it as a not-before floor on the gap.
   • Articles per run — left as 1 for the simplified Hopper model.
   • Monthly cap — soft self-imposed ceiling on briefs per month. 0 = unlimited (your tier's monthly article cap still applies).
   • Approval mode — Manual = emitted briefs land as drafts for your review; Auto = published immediately as drafts ready for the tick pipeline.
   • Auto-generate articles — off by default. See Auto-generate & auto-publish below.
   • Final article state — shown when Auto-generate is on. Draft (you review before publishing) or Published (live immediately).
5. Click Save & Close. The Hopper appears in the list.

Step 2 — Generate & curate candidates

Open the Hopper you just created and switch to the Candidates tab.

Generate the first batch

On a brand-new Hopper the tab shows a centred “No candidates yet” card with a Generate candidates button. Click it. What happens behind the scenes:

1. Harvest — the watch type is dispatched. Search queries hit your configured search provider (Serper / Tavily / Google); RSS feeds are fetched and parsed; URL sets are fetched directly. Search queries also try a Serper news endpoint variant for journalist sources.
2. Dedupe — URLs are deduped by host+path so the same article doesn't enter twice.
3. Inline verification — each candidate URL is fetched (5s timeout per URL, 45s overall budget) and the body checked. The verifier requires the page to actually load, be long enough to be a real article, and contain at least 2 keywords drawn from the search title+snippet. URLs that fail are marked Bad. URLs whose verification timed out are marked Unverified.
4. Persist — rows land in the Candidates tab already badged Good / Bad / Unverified.

The Run Now message after harvest tells you the breakdown: e.g. “Ran 1 Hopper(s) — 21 new candidate(s) found, 1 brief(s) emitted. (3 source(s) dropped by inline verifier — see the Candidates tab.)”

Curate the list

Each row carries a Status badge:

• Good — in the source list. Every Run Now / scheduled run uses this URL as a source.
• Bad — excluded. Verifier dropped it (or admin marked it Bad). The row stays so you can audit.
• Unverified — harvest budget ran out before checking. Click Verify on the toolbar to clear the residue.

Toolbar buttons:

• Run Now — produce ONE brief from the current Good list. See the next section.
• Refresh candidates — harvest again. Duplicates skip on signature; new URLs land verified.
• Create brief from selected sources — tick rows then click. Literal selection, no filter. One brief with exactly the URLs you ticked as sources.
• Always include — promotes the URLs of selected rows to the Hopper's default-sources list, so every future emitted brief inherits them.
• Mark good / Mark bad — admin override on the verifier verdict. Use when you know the URL is good but the verifier was over-strict (or vice-versa).
• Delete — hard delete of the row. Confirmation appears inline (Joomla framework dialog — no browser modals). The URL signature is kept, so a future Refresh could rediscover the URL.
• Verify (N) — runs the verifier across Unverified rows. Visible only when there are unverified rows to act on.

Click row titles to open the source URL in a new tab. Hover the snippet for the full text.

Step 3 — Run Now: the first brief

When you have at least one Good row, click Run Now on the Candidates tab.

The Hopper:

1. Loads every Good row's URL.
2. Calls the AI to synthesise a title, topic statement, and keyword list from the cluster.
3. Creates ONE brief whose sources field includes those URLs plus the Hopper's Always-include source URLs.
4. Stamps the Hopper's last_run so the cadence doesn't fire again immediately.
5. Redirects you to the new brief's edit screen.

From there, click Generate Article on the brief's toolbar to run the article-gen pipeline. The pipeline researches across all the source URLs and writes a piece that cross-references them — not a paraphrase of any single source.

Step 4 — Cadence & time of day

Manual cadence (the default) is fine for one-off topics. For ongoing coverage, set a cadence and a time of day.

Cadence options

How the time of day works

The time-of-day field is the “earliest the Hopper becomes eligible to fire today”. The scheduled task checks each Hopper on every tick:

1. If today's site-timezone clock hasn't reached the configured HH:MM, the Hopper isn't due. Wait.
2. Once the time of day has passed AND the last run is older than the cadence's minimum gap, the Hopper fires on that tick.

So a Hopper set to Daily at 02:00:

• The 01:00 cron tick: not due (haven't reached today's 02:00 target yet).
• The 02:00 cron tick: due (we're at the target and last run was ≥23h ago). Fires.
• The 02:05 cron tick: not due (just ran, <23h gap).

If your cron runs less frequently than the time-of-day target — say once daily at 05:00 — the 02:00 Hopper still fires at 05:00 (it's the next tick after the target). The cadence drift problem of pure gap-based scheduling is gone, but you do still need the scheduled task to be running often enough to actually hit your target window.

Step 5 — Scheduled task setup

The unified Article Generator scheduled task drives every Hopper's cadence (plus on-demand batch jobs and the unattended generation phases). Create it once and leave it running.

Create the task

1. Go to System > Scheduled Tasks > New.
2. Pick task type Article Generator.
3. Give it a name (e.g. “Article Generator”).
4. On the Execution Rule tab, choose Interval, in minutes, set to 5. This is the recommended cadence: it keeps the Hopper time-of-day accurate to within 5 minutes, and the auto-generate phase ticks the article-gen pipeline forward 20 briefs per run.
5. On the Options tab, set Status to Enabled.
6. Save.

What the task does on each tick

1. Hopper discovery (30s budget) — runs every Hopper whose cadence is due. Calls produceBriefs; new briefs land as drafts (or as state=1 if the Hopper's Approval mode is Auto).
2. Auto-generation (90s budget, max 20 briefs) — for briefs whose Hopper has Auto-generate articles on, ticks the article-gen pipeline forward one step. The same step machine the browser drives, headless. A brief that takes 5 ticks to finish will progress one step per scheduled tick.
3. Auto-publish (30s budget) — briefs that finished generation with Final article state = Published get their Joomla article flipped to state=1.
4. Batch queue (30s budget) — services on-demand jobs from the Generate All Steps button.

Logs

Each task's log lives at administrator/logs/task_{id}.log.php. Phase-level events also mirror to administrator/logs/com_artigen.error.php. A healthy run looks like:

2026-06-25T02:05:01+00:00  INFO  task9  Running task#09 'Article Generator'.
2026-06-25T02:05:01+00:00  INFO  task9  Task> Hopper #1 "AI ethics watch": 4 source(s) used, 1 brief(s) emitted
2026-06-25T02:05:01+00:00  INFO  task9  Task> Hopper discovery phase: 3 due, ran 1, skipped 2 (not yet due)
2026-06-25T02:05:02+00:00  INFO  task9  Task> Auto-generation phase: ticked 1 brief(s), 0 completed, 0 failed
2026-06-25T02:05:02+00:00  INFO  task9  Task> Auto-publish phase — no completed briefs awaiting publish
2026-06-25T02:05:02+00:00  INFO  task9  Task> Batch queue phase — no pending batch jobs
2026-06-25T02:05:02+00:00  INFO  task9  Successfully finished task#09 in 0.14 seconds.

Step 6 — Auto-generate & auto-publish (optional)

If you want the loop fully unattended, switch on the two automation knobs on the Hopper edit form (Production tab):

• Auto-generate articles — when on, briefs this Hopper emits are picked up by the scheduled task's auto-generation phase. Each cron tick advances a brief one step. A typical article finishes in 5–7 ticks (5 to 35 minutes wall-clock at the recommended 5-minute cadence).
• Final article state —
  • Draft — article generates fully but stays unpublished. You review and click Publish in Joomla's article editor. Recommended for new Hoppers.
  • Published — article goes live as soon as the pipeline completes. Use only when you trust the Hopper's source curation completely; the article never gets a human review pass.

Briefs inherit the Hopper's auto_publish_state at emit time, so changing the Hopper's setting only affects briefs emitted from that point on. Previously-emitted briefs keep whatever was set when they were created.

Troubleshooting

Full documentation index: Article Generator documentation. Changelog: artigen-changelog.html.