Fresh Toggle, a Recency Weight Slider, and Much More Detailed AI Hints Answers
Recency becomes something the visitor asks for rather than something you decide for them, and the AI answer above the results stops summarising one document and starts reading all of them.
01 A Fresh toggle, owned by the visitor
The hosted search page now carries a Fresh toggle beside the sort controls. Turning it on pushes recent documents up the ranking: with the default settings a page published today counts about twice as much as the same page a year old.
It only changes the ORDER. Nothing is hidden, the result count is identical, and documents keep the place their relevance earns them. It is off by default, and the choice rides in the URL, so a fresh-biased search can be bookmarked and shared, and it survives paging and facet clicks.
Fresh is available on the hosted search page, in the Drupal module and the WordPress plugin, and as a fresh_bias option in the LangChain, LlamaIndex, Haystack, MCP and Laravel packages.
02 How hard it pushes is yours to set
Whether recency counts stays the visitor's choice, but a new Fresh Results Bias Weight slider in Search Tuning decides how much it counts when they turn it on.
| Weight | Effect on ranking |
|---|---|
0.1 | A newer page only wins ties. |
1.0 | The publication date all but replaces relevance — the newest match comes first almost regardless of how well it matches. |
The slider tells you what your value means in plain words — „half its boost after about 2 days“ — so you are not guessing at a number. It is set per index, sits in the same place on the Drupal and WordPress admin screens, and is configurable from code in the LangChain, LlamaIndex, Haystack, MCP and Laravel Scout packages.
The curve behind it was retuned
The old recency curve halved a document's boost only after a year, so on a site publishing weekly the toggle reordered almost nothing a reader would notice. The default now halves it after about ten days, and the weight slider takes it as far as six hours if your content moves that fast.
Undated pages no longer jump the queue
A page with no publication date was being treated as if it had been published the very instant you searched, which put undated pages at the top of a „newest first“ list — the opposite of what the toggle promises. Pages without a date are now left out while Fresh is on. Fixed everywhere at once: the hosted search page, the Drupal module, the WordPress plugin and all five packages.
03 AI Hints answers read everything relevant
AI Hints answers used to be two or three generic sentences drawn from whichever result ranked first. They are now written in Markdown — a direct opening sentence, then the specifics as a list — and they draw on every retrieved document that bears on your question, naming the people, places, dates and numbers involved.
A short, generic summary of the top-ranked result, hedged when unsure.
One coherent answer combining what several articles say — and a plain „no answer in this index“ when there genuinely is none.
04 Full text in Drupal and WordPress, and the same answer everywhere
- Full text, not teasers. AI Hints in the Drupal module and the WordPress plugin were being shown only the short teaser lines already visible on the results page — and on many sites nothing at all — so the assistant answered from almost no material. It now reads the full text of the most relevant matches straight from your index, including PDFs and Office documents added through the Data Ingestion API, which were previously filtered out of the answer even though they appeared in the results.
- One prompt, checked automatically. The hosted search page, the RAG sandbox, the Drupal module, the WordPress plugin and all five client packages now build the prompt and the retrieved context identically, verified against a shared reference rather than kept in step by hand. A question asked through the API and the same question typed into your search page produce the same answer.
- A comfortable reading column. The Document Reader text was running the full width of the browser, which on a wide monitor meant very long lines. It now uses the same reading column the Drupal and WordPress readers already used.
05 Queries matching pages with code samples no longer error
Searches whose results included a page containing code samples returned a blank error page instead of the results. The AI Hints answer passes the matched content to the model, and certain snippets in that content were being misread while the page was assembled. Result text is now handed over exactly as indexed — punctuation, quotes, accents and code all preserved character for character — so any query works regardless of what the matching pages contain.