Domain-Specific Search: Why Generic Search Fails Specialist Catalogues

A buyer types the exact name of a compound you stock. Your search decides it is a typo, changes it to a common English word and shows them nothing useful.

The short answer: domain-specific search is search that learns what a valid word is from your own catalogue, not from general language. Search left on general-language defaults can, on a specialist catalogue, "correct" valid industry terms into common words, drop short codes, split part numbers and miss the second name for the same product. The fix is to correct spelling against your catalogue's own vocabulary, test with your own failed searches, map the abbreviations and alternative names your buyers use, and set rules for the items that matter most.

For heads of digital and product in pharma, industrial supply, chemicals, research and B2B distribution: where generic search breaks on your terms, a vocabulary audit to run this month, and what to fix first. This post is about the words themselves; how people phrase searches is covered in improving enterprise search for real-world queries.

Why does generic search get industry terms wrong?

A search engine has to decide three things about every word a buyer types: is it spelt right, is it worth indexing, and what else means the same thing. Many search setups answer at least one of them from general language: a spell-checker built on a dictionary, a language-wide list of stop-words and word endings, or default rules for what counts as a word. Even engines that correct towards your own indexed words still apply general rules for word length, capital letters and punctuation.

• Spelling. Where spelling correction uses a general dictionary, most compound, drug, alloy or species names look like typos and are changed to the nearest common word (how a general dictionary mis-corrects specialist terms). Engines that correct against your own catalogue avoid this, so it is the first thing to test.
• What counts as a word. Some engines skip very short words by default, as the MySQL manual's full-text settings show (MySQL, "Fine-Tuning MySQL Full-Text Search"; see also short terms missing from the index). In a technical catalogue, a two-character term is often the whole query: pH, O2, Cu, PE.
• What means the same thing. A general synonym list knows that "car" and "automobile" match. It does not know that your buyers say "PTFE" where your catalogue says "polytetrafluoroethylene".

The failure is silent: the buyer concludes you do not stock the item.

What is domain-specific search?

Domain-specific search is search that takes its idea of a valid word from your catalogue and your field rather than from a general dictionary. In practice it does four things a generic setup does not:

1. Corrects spelling against your own terms. A misspelling is matched to the nearest word that appears in your catalogue, not in everyday English.
2. Keeps codes, units and short terms intact. Part numbers, grades, concentrations and two-letter terms are indexed and matched as they are written.
3. Knows your alternative names. Abbreviations, trade and generic names, older names and regional spellings are mapped to the same items.
4. Lets you set rules for key items. For searches that matter commercially, you decide which item comes first and what should never appear.

Which kinds of words cause the most trouble?

Abbreviations that mean different things

Technical buyers abbreviate constantly, and many abbreviations collide with other meanings: "PE" can be polyethylene, a price-earnings ratio or physical education. In medicine the risk is serious enough that the Institute for Safe Medication Practices publishes a list of abbreviations that have been misread in medication errors (ISMP, "List of Error-Prone Abbreviations").

Units and capital letters

Unit symbols carry meaning in their case: under the SI rules NIST sets out, the mega prefix "M" is upper case and milli "m" is lower case (NIST, Guide to the SI, chapter 6). Many search setups lower-case every word to be forgiving, so "mS" (millisiemens) and "MS" (megasiemens, or mass spectrometry) become the same token. Numbers add their own variations: "10mm", "10 mm" and "10-mm" may be indexed three different ways.

Part numbers and codes

Product codes mix letters, digits, hyphens and slashes. A default setup may split a code at the hyphen and match either half on its own, so a search for one part returns every item sharing a fragment. Buyers also type codes without the separators.

Chemical names

One substance can have generic, systematic, common and trade names; CAS assigns each substance a single Registry Number precisely because the names alone cause confusion (CAS, "CAS Registry"). A buyer may search by any of them, including the number.

Brand names and generic names

Medicines have a generic name and often several brand names. The WHO assigns each active ingredient a unique International Nonproprietary Name that anyone may use (WHO, "International Nonproprietary Names"), but buyers often remember the brand. The same ingredient can also carry different generic names in different countries, as paracetamol and acetaminophen do. Industrial catalogues have the same pattern with manufacturer part names.

Other languages and spellings

Aluminium and aluminum, sulphur and sulfur, and German compound nouns that join several words into one. Elastic's documentation demonstrates this with "Donaudampfschiff", which needs a word list to split into its parts before each part is searchable (Elastic, "Dictionary decompounder token filter"). A catalogue sold in several countries has to treat each version as the same thing.

The vocabulary audit: a table to run on your own catalogue

One row per term type: an example, what typically goes wrong in a generic setup, and the fix. The examples are illustrations, not findings from any customer's catalogue; replace them with terms from your own data.

Term type	Example (illustrative)	What goes wrong in generic search	How to fix it
Specialist term not in a general dictionary	A compound, species or alloy grade name	"Corrected" into a common word; results for the wrong thing	Correct spelling against your catalogue's own vocabulary, not a general dictionary
Abbreviation or acronym	PTFE for polytetrafluoroethylene; PE for polyethylene	No match to the full name, or a match to an everyday meaning	Map each abbreviation to its full form from your catalogue and query logs
Very short term	pH, O2, Cu	Not indexed at all, so the query can never succeed	Index short terms; test each one correctly spelt
Unit with case meaning	mS vs MS; mg vs Mg	Lower-casing merges two different things	Keep case in unit and code fields, or map the cases to separate terms
Number with unit	10mm, 10 mm, 10-mm	Indexed as different tokens; buyers miss items written another way	Normalise number-and-unit forms at indexing and at search
Part number or code	AB-1234/X typed as AB1234X	Split at punctuation; fragments match unrelated items; joined form not found	Index codes both joined and split; rank exact code matches first
Chemical identity	Ethanol, ethyl alcohol, CAS 64-17-5	Only one name works	Carry the alternative names and the registry number with the item
Brand vs generic	Paracetamol vs acetaminophen vs a brand name	Buyer's name does not match the catalogue's name	Map brand, generic and regional names to the same items
Spelling variant or language	Aluminium/aluminum; a German compound noun	One spelling or language finds nothing	Map variants; split compounds where your catalogue uses them
Plural with a different meaning	Glass (material) vs glasses (eyewear)	General word-ending rules merge them	Check the top plural forms in your queries; exempt the ones that change meaning

How to run the audit

1. Pull the words that matter. Export the catalogue and a month of search queries, including those that returned nothing.
2. Sort them by term type. Use the rows above. Many catalogues are dominated by two or three types.
3. Search each one correctly spelt. Note whether the engine changes it, splits it or ignores it.
4. Replay your failed searches. Mark each zero-result query whose answer exists in the catalogue. That count is your vocabulary gap.
5. Fix in order of traffic, then keep the list. Fix the term types most searches hit first and re-run the list after every change.

A fuller method for building a test set, including misspelt and reordered versions of real items and nonsense queries that must return nothing, is in a realistic dataset for comparing enterprise search solutions.

How do you fix search for a specialist catalogue?

Work in order of how many searches each problem affects, not in order of what is easiest to configure.

Correct spelling against your own catalogue

This has the widest effect. When the engine only corrects towards words in your catalogue, a valid technical term stays as typed, a mistyped one moves towards a product you sell, and the vocabulary keeps up as the catalogue changes.

Map the alternative names buyers use

Abbreviations, brand and generic names, old product names and regional spellings need an explicit list built from your catalogue and query logs. Keep it small and checked: every mapping is a decision about what a word means in your field.

Test with your own failed searches

Your zero-result report is your best source of vocabulary problems. Read the top no-result searches monthly, sort them by term type, and add each fixed one to the audit list so it stays fixed.

Set rules for the items that matter most

For a few searches, nearly right is not good enough: the flagship product, the item a key customer reorders weekly, the regulated product that must not appear for a near-miss. Pin the right item to the top for those searches and hide what should never appear, then compare results with and without your rules before you publish them.

What can you fix yourself, and what needs expert help?

ITMTB builds and tunes its own search engine around a catalogue's vocabulary, and it is available two ways.

Self-serve on Tomorrow Central. Tomorrow Central Search takes a catalogue as a CSV, JSON or Google Merchant product feed file, or imports it from your website once the site is verified. These parts of the vocabulary work you can do yourself:

• Spelling is corrected against your own catalogue's words, not a general dictionary, and each search project has its own vocabulary.
• The statistics show your top no-result searches and how often spelling was corrected, so you can find the terms that fail and replay them. Your own Try-it searches are not counted in the statistics.
• You can pin or hide results for specific searches and preview results with and without your rules.

Going live needs one proof that you own the website (a DNS record or a small file on the site); uploading and testing do not. There is a free plan (at the time of writing: up to 5,000 products, 10,000 searches a month and one search project), so a catalogue of that size, or a 5,000-item sample of a larger one, can be audited at no cost. Setup is covered in Tomorrow Central's guide to site search for a catalogue.

On one live catalogue, the same engine cut zero-result searches roughly tenfold (measured differently before and after; see the case study).

What is not self-serve today. You cannot yet edit your own list of synonyms and abbreviations in Tomorrow Central Search. Mapping PTFE to polytetrafluoroethylene, or a brand name to its generic, is tuning work ITMTB does when it deploys the engine for you. Handling codes, units and capital letters the way your catalogue writes them is part of the same tuning work.

On your own infrastructure. For organisations that need the engine inside their own environment, ITMTB deploys and tunes it there, including the vocabulary mapping. See enterprise search. If you are still deciding whether to build, buy or use a managed engine, build vs buy enterprise search compares the three.

What usually goes wrong when teams try to fix this?

1. Mapping a term once for the whole catalogue. PE means polyethylene in plastics and something else in finance; scope each mapping.
2. Turning spelling correction off. Valid terms stop being changed, and real typos stop being fixed. Correct against your catalogue instead.
3. Lower-casing every field. Keep case in unit and code fields.
4. Indexing codes only in split form. Buyers type them joined.
5. Applying a general plural rule everywhere. Exempt the plurals that change meaning.

The general mistakes in search projects are listed in what usually goes wrong.

Frequently asked questions about domain-specific search

What is domain-specific search?

Search that takes its idea of a valid word from your own catalogue rather than general language. It corrects misspellings against your catalogue's terms, keeps codes and units intact and maps the alternative names buyers use.

Why does site search fail on industry terms?

Many setups rely on general-language defaults: a dictionary-based spell-checker treats a valid technical term as a typo and changes it, short codes can be dropped from the index, and general synonym lists do not know that two names in your field mean the same thing.

Can a general synonym list fix specialist vocabulary?

Rarely. A general list does not know your field's abbreviations or trade names and can add wrong everyday matches. Synonyms for a specialist catalogue need to come from that catalogue and its buyers' queries.

How do I test whether my search understands our vocabulary?

Search for your own specialist terms spelt correctly and check that none is changed into another word. Then replay a month of zero-result searches and count how many have a matching item in the catalogue. Both lists come from your own data.

Is AI or semantic search the answer for technical catalogues?

It can help with descriptive questions, but it does not replace exact handling of codes, units and part numbers, and it tends to return something for every query. Test it with the same audit, plus nonsense queries that should return nothing.

Does domain-specific search need a big search team?

No. Most of the work is a vocabulary audit, rules for key items and a monthly look at the no-result report; the tuning can be done by a specialist.

Key takeaways

• Search left on general-language defaults can change valid terms, drop short codes, split part numbers and miss alternative names on a specialist catalogue.
• Domain-specific search takes its vocabulary from your catalogue: it corrects spelling towards your products and keeps codes, units and abbreviations intact.
• Run the vocabulary audit, one row per term type, against correctly spelt terms and your real failed searches, and re-run it after every catalogue change.
• Fix in order of traffic: catalogue-based spelling correction, then alternative names, then rules for key items.

Where to start

To run the audit on your own catalogue, sign up for Tomorrow Central Search on the free plan, upload the catalogue or a 5,000-item sample, and search your specialist terms with the signed-in Try-it search before anything goes live. If you need abbreviation and synonym mapping for your field, or the engine inside your own infrastructure, see enterprise search or contact us.

Sources

• MySQL, "Fine-Tuning MySQL Full-Text Search", MySQL 8.4 Reference Manual, accessed 19 October 2026: https://dev.mysql.com/doc/refman/8.4/en/fulltext-fine-tuning.html
• Institute for Safe Medication Practices, "List of Error-Prone Abbreviations, Symbols, and Dose Designations", accessed 19 October 2026: https://home.ecri.org/blogs/ismp-resources/list-of-error-prone-abbreviations
• NIST, "NIST Guide to the SI, Chapter 6: Rules and Style Conventions for Printing and Using Units", Special Publication 811, accessed 19 October 2026: https://www.nist.gov/pml/special-publication-811/nist-guide-si-chapter-6-rules-and-style-conventions-printing-and-using
• CAS, "CAS Registry", accessed 19 October 2026: https://www.cas.org/cas-data/cas-registry
• World Health Organization, "International Nonproprietary Names (INN)", accessed 19 October 2026: https://www.who.int/teams/health-product-and-policy-standards/inn
• Elastic, "Dictionary decompounder token filter", Elasticsearch documentation, accessed 19 October 2026: https://www.elastic.co/docs/reference/text-analysis/analysis-dict-decomp-tokenfilter
• ITMTB enterprise search case study: /blog/enterprise-search-case-study-business-research-company