Live Web Search for Fresh Topics
What it does: Lets the AI search the internet before writing, so you can publish about news and events from hours ago — not just what the model learned in training.
Applies to Antradus AI Publisher v2.8.0 · Last updated 2026-08-28
When to use it
- Breaking news, product launches, sports results, anything from the last days/weeks.
- Topics where being current matters more than speed or cost.
Search alone is not proof of freshness. The model decides what to search for and what to believe,
and nothing checks the dates on what it finds. When the date genuinely has to be right, use
Google Trends instead: it fetches each source, reads the
publication time out of the page, and refuses anything that is not from today. Web search still runs
there — it just is not the only thing standing between you and a stale fact.
When your topic is evergreen ("how to compost"), leave it off — it adds time and cost for no benefit.
Step by step
- In the Antradus AI panel, tick 🌐 Search the web (for news & recent events).
- Write your topic as you normally would — being specific about recency helps: "what changed in today’s EU AI Act vote".
- Start Writing. The AI searches first, reads what it finds, then writes with fresh facts.
Provider support (important)
| Provider | Web search |
|---|---|
| OpenAI | ✅ — needs a search-capable model (e.g. gpt-5, gpt-4o) |
| Anthropic (Claude) | ✅ |
| Google Gemini | ✅ (Google Search grounding) |
| OpenRouter | ✅ on any model — including free ones (though some models use results better than others) |
| DeepSeek | ❌ not supported by DeepSeek’s API |
Cost & speed
- Web search is billed by your AI provider on top of normal token costs (each has its own search pricing). The plugin adds no fees of its own.
- Expect generation to take noticeably longer — the AI is reading the web before writing.
Tips & gotchas
- Works everywhere articles are generated — including Bulk Publishing (there’s a web-search option on bulk batches too).
- Clean articles, no citation code (v2.0.2): some search-capable models sneak raw citation snippets like
([site.com](https://…utm_source=openai))into the text. The plugin now cleans these automatically — stray markdown links become normal links, bare source stubs are removed, and AI tracking parameters are stripped from URLs. - Comparisons research every subject (v2.0.2): for topics like "X vs Y vs Z", the AI is instructed to search each named product or model separately and cover all of them — and to say so explicitly when one of them doesn’t offer the capability being compared. It also always knows today’s date, so articles are framed in the present.
- Today’s products, not remembered ones (v2.0.4): the AI must first search for what is genuinely current — the newest generation, version, price, or ranking of any subject — and centre the article on what the search returns; things it merely remembers as "the latest" may only appear as background. On OpenAI, searching is mandatory when the option is ticked, so the model can never skip it and write from stale memory.
- If results seem stale on OpenRouter, try a different model — some ignore search context.
- Combine with a source URL for the best of both: the source anchors the story, the search adds the latest developments.