Concrete cases, technical observations, and lessons learned, published monthly.
The code of the three connectors, re-read on 21 October 2026: one hook, no template rewritten, no table created, and the original search taking over when the engine does not answer.
Read the article →Five pages read across six sites: almost everything you admire at a competitor’s is laid down by their platform, and the only gap between the five is an absence.
Read the article →Five product pages from French distributors, read on 20 October 2026: what a machine finds there, what it does not, and four moves that cost no advertising.
Read the article →Both ordering notations return the same annotations — because the engine reads neither. Measured on the electrical pack, with what it does to ranking.
Read the article →15x21, DN15, half-inch: one part, three sizings. The plumbing pack isolates two, loses the quote-mark inch, and produces a pipe annotation that does not exist.
Read the article →Who publishes what, which one binds member countries, and three prefixes out of four the engine can turn into an attribute. The fourth one, it does not read.
Read the article →An exact SKU returns 0 results. “M8-30” returns 1 product out of the 5 that match, and redirects the customer to it. A typo returns 0. Measured on a WooCommerce shop, plugin disabled.
Read the article →The typo is found: our page said otherwise. The real flaw is elsewhere. M8 is not indexed, and one typo in a single product turns the correction off for every visitor.
Read the article →Three MCP tools, 12 natural-language queries, one catalog indexed with and without a pack. What the agent receives, what it misses, and what it cannot discover.
Read the article →The whole product in one call, or identifiers only. The engine never composes a URL; the front end does. And allowed_origins, measured.
Read the article →A pack almost always has a higher average order value, but naturally loses in pure relevance ranking against a single product. Rather than deciding inside the score, a separate field.
Read the article →In-house development seems free on paper. What it actually costs, and where Heurix sits on this trade-off.
Read the article →Your catalog is indexed and search works. What comes after: reading what customers search for, fixing rankings.
Read the article →A category page has no query to process — just products to rank well. Index, verify, render.
Read the article →Ask a natural-language question about your catalog directly inside your AI agent — fifteen minutes to install.
Read the article →Semantic similarity is a real advance, and the wrong tool when references encode dimensions and standards.
Read the article →An optimal algorithm on 500 products becomes a bottleneck on 50,000, and the reverse is just as true. Why the right answer comes from measurement.
Read the article →A dashboard always shows more numbers than you have decisions to make. Here are the three worth your attention each week, and what they actually trigger.
Read the article →Sales, Black Friday, a new range: traffic rises, and with it the cost of every search that fails. What to check before, during and after.
Read the article →What share of your revenue goes through the search bar, and what each failed search costs you. The calculation method, variable by variable.
Read the article →You're running a campaign on a range. The engine knows nothing about your promotion: it ranks by relevance. Here's how to take control of a few searches without upsetting everything else.
Read the article →A customer looks for a product you have in stock, and gets nothing. They won't write to tell you. Here's how to spot those searches, understand why they fail, and fix them.
Read the article →The root of the name doesn't mean “to search.” It means “to find.” The etymology, honestly told.
Read the article →Two API calls, one real search result with a typo and a different format. All code is tested.
Read the article →The honest inventory of what you'd have to write and maintain yourself — analyzers, tokenizers, synonym graphs.
Read the article →Five tools, five different approaches to finding the right product fast — where each one factually excels.
Read the article →A rule pack covers the essentials of an industry, but never your own catalog's exact vocabulary.
Read the article →What their documentation says about a technical reference, and what our engine still gets wrong.
Read the article →A customer who searches and finds nothing tells you, in their own words, what they wanted to buy.
Read the article →Product search gets configured with care, while blog or FAQ search runs on CMS defaults.
Read the article →The listed price doesn't say much until you know what it actually bills.
Read the article →Why default engines fail on technical product codes, and how regex changes the game.
Read the article →Indexing limits of large catalogs, and how to fix it without rebuilding the site.
Read the article →Designing dynamic, well-ordered facets for technical catalogs.
Read the article →Designing a business synonym dictionary without creating false positives.
Read the article →The economic leverage of internal search, and a simple method to size the current loss.
Read the article →The three most common blind spots in default search engines, and how to spot them in five minutes.
Read the article →Concrete symptoms to check on your own site, no technical tools required, in under ten minutes.
Read the article →Synonyms, facets, typo tolerance: what these terms actually mean for your catalog.
Read the article →