How to Conduct a Prior Art Search: A Researcher-Friendly Workflow for India

An assistant professor brings in a solar crop dryer her final-year students built for chilli growers. "We searched Google and Google Scholar," she says. "There's nothing like it." Half an hour on Espacenet turns up a Chinese utility model with the same fan arrangement and an Indian journal paper that tested almost the same chamber three years earlier.

The gap between a quick search and a real one is what this guide is about. It explains how to conduct a prior art search before you file or brief a professional: what Indian law counts as prior art, an eight-step workflow, the databases and syntax that work, a worked example, and how to record what you did.

Quick answer

To conduct a prior art search, write your invention as a short list of features, turn each feature into keywords and synonyms, find the matching IPC and CPC classification codes, and run combined keyword-and-code searches in InPASS, Espacenet, Google Patents and PATENTSCOPE, plus journals and Indian theses. Follow the citations of the closest hits, map strong documents against your features in a table, and stop when new searches return only documents you have seen. Record every database, string and date. Under the Patents Act, 1970, prior art is anything made public anywhere in the world, in any language, before your priority date, including your own papers.

What counts as prior art under Indian law

Section 2(1)(j) of the Patents Act, 1970 defines an invention as a new product or process involving an inventive step and capable of industrial application. Section 2(1)(l) defines a "new invention" as one not anticipated by publication in any document, or used in the country or elsewhere in the world, before the filing date, so that it has not fallen into the public domain or become part of the state of the art.

Three things follow:

• Worldwide and in any language. A Chinese utility model or a German journal article counts as much as an Indian patent.
• Any form of disclosure. Patents, papers, posters, theses, manuals, brochures, YouTube demonstrations and public sales all count. In many fields the closest art is non-patent literature.
• Your own work counts against you. Your paper, your tech-fest poster, your M.Tech thesis on Shodhganga and your PhD synopsis on Shodhgangotri are all prior art once public.

Sections 29 to 34 carve out narrow exceptions: matter published without your consent (section 29), communication to the Government for investigation (section 30), display at a notified exhibition or a paper read before a learned society, if you file within twelve months (section 31, now claimed on Form 31 under rule 29A, added in 2024), and public working for reasonable trial within a year before the priority date (section 32). Patent Offices have applied the learned-society exception unevenly, so treat these as rescue routes, not a plan. Our guide on patenting an idea without disclosing it too early sets out the safer sequence.

Two blind spots matter. Indian applications are not ordinarily published until 18 months from the filing or priority date, whichever is earlier (rule 24 of the Patents Rules, 2003), so no search sees the newest filings. And an Indian application filed before yours but published later can still be cited for prior claiming under section 13(1)(b). Novelty is also only half the test: two known documents combined can defeat inventive step, as our guide on what can be patented in India explains.

Steps 1 to 3: features, keywords and classification codes

Step 1: write the invention as features

Before you open a database, write one claim-like sentence and break it into four to seven features. Use plain technical words, not your lab's nickname for the device. Mark the one or two features you believe are new. This list drives the keywords, the codes and, later, the chart that shows whether a document hurts.

Step 2: keywords, synonyms and spellings

For each feature, list how other people would describe it:

• Synonyms and broader terms: dryer, drier, dehydrator; fan, blower, ventilator.
• Spelling variants: aluminium and aluminum, sulphur and sulfur, chilli, chili and chile.
• Abbreviations: PV, RH, IoT, PLC.
• Patent drafting language: "drying enclosure", "air heating means", "control unit configured to".

Step 3: classification codes

Keywords miss documents that use different words or another language. Classification codes don't. The IPC is used by patent offices worldwide, including India; the CPC, managed jointly by the EPO and the USPTO, subdivides it much further. Find codes in WIPO's IPCPUB and the Espacenet CPC browser, or read the codes on the three closest documents from a keyword search. Read the scheme's notes too. Groups in curly brackets in the CPC scheme are CPC-only and will not work in InPASS, which searches IPC.

Step 4: where to search, and what each database is good for

Use at least one Indian source, one worldwide patent database with good classification search, and your field's literature.

A word on TKDL. Full access is limited to patent offices, including the Indian Patent Office, under a non-disclosure agreement; the public site offers only a representative database of about 1,250 formulations. For a herbal or Ayurvedic composition, assume the examiner can see classical texts you cannot, and remember that section 3(p) excludes traditional knowledge and aggregations of its known properties.

Step 5: search strings that actually work

Join the synonyms for each feature with OR, join the features with AND, add proximity where supported, then AND the result with a classification code. Each database writes this differently:

• Espacenet: fields such as ta (title or abstract), ctxt (title, abstract or claims), ftxt (all text), cpc= and ipc=, plus any and all for word lists; truncation with * (any characters), ? (none or one) and # (exactly one); proximity with prox/distance<n.
• Google Patents: Advanced Search, with TI=, AB=, CL= and CPC= fields and NEAR/x, ADJ/x, WITH and SAME. Google's help says proximity changes ranking, not which documents are retrieved, and wildcards work on English words only.
• PATENTSCOPE: EN_TI, EN_AB, EN_CL, EN_ALLTXT, IC (subgroups included) and CPC; proximity as "word1 word2"~n. No wildcard at the start of a word, and three wildcards per query unless you log in.
• InPASS: AND, OR and NOT between rows, quotes for phrases, * for any characters and ~ for one character, so dr~er catches dryer and drier. No proximity operator, and IPC codes only.
• USPTO Patent Public Search: ADJn, NEARn, WITH and SAME, with field tags such as .ti., .clm. and .cpc.

Three strings from the worked example below show the pattern:

Espacenet: ta=solar AND ta any "dryer drier dehydrator" AND (humidity prox/distance<5 fan) AND (cpc="F26B3/28" OR cpc="F26B3/286")

Google Patents: ((humidity OR moisture) NEAR/10 (fan OR blower)) AND (dryer OR drier OR dehydrator) AND F26B3/28

PATENTSCOPE: EN_ALLTXT:(("solar dryer"~5 OR "solar drier"~5) AND "humidity fan"~10) AND IC:F26B3/28

Because NEAR only ranks in Google Patents, that string retrieves every dryer document in the class mentioning humidity and a fan anywhere, close matches first. Read well beyond the top of the list.

Steps 6 and 7: citation chaining, screening and the claim chart

Step 6: follow the citations

Mine every close document. Go backward to what it cites, including examiner citations; European and PCT search reports mark the most relevant as X (relevant alone) or Y (relevant in combination). Go forward to later documents citing it, through Espacenet's citing documents or "Cited by" in Google Patents and Google Scholar. Check the family too: a version filed elsewhere may have an English text or clearer drawings.

Step 7: screen in passes, then chart

1. Title, abstract and main drawing: discard the clearly irrelevant.
2. Claims and figures of what is left.
3. The shortlist in full. A document discloses more than its claims.

Then map the strongest five to ten documents against your features, as in the example below. The chart shows whether one document has every feature (a novelty problem) or two together cover them (an inventive step problem).

Worked example: a low-cost solar crop dryer

Take a dryer for onion slices and chillies designed at an agricultural engineering college in Gujarat:

• F1: a drying chamber with perforated trays;
• F2: a solar air heater feeding warm air to the chamber;
• F3: a DC fan run from a small PV panel, with no grid connection;
• F4: a humidity sensor at the exhaust that controls fan speed;
• F5 (believed new): a damper that recirculates part of the exhaust air when the inlet-outlet humidity difference falls below a set value, saving heat late in the day.

Keywords: solar, photovoltaic, PV; dryer, drier, dehydrator; chilli, chili, onion, produce; fan, blower; humidity, moisture, RH; recirculat*, damper.

Example codes (checked against IPC 2026.01 and CPC 2026.08; confirm the current version before relying on them):

• F26B 3/28 (IPC and CPC): drying by radiation, e.g. from the sun.
• F26B 3/286 (CPC only): solar radiation combined with convection.
• F26B 25/22 (IPC and CPC): controlling drying according to the material's liquid content.
• A23B 7/02: dehydrating fruit or vegetables. CPC flags this group as affected by a reclassification, so follow its warning note to the newer groups.
• Y02P 60/85 (CPC tagging code): food storage or conservation, e.g. cooling or drying.

Run the three strings, then InPASS with solar dr~er in the complete specification field and the F26B codes in the IPC field, then Google Scholar and Shodhganga. Suppose the shortlist looks like this (illustrative documents, not real citations):

The judgement: no single document has all five features, so the device looks new. But F1 to F4 are known, and D2 with D3 comes close to F5. The real question is inventive step: would a skilled engineer reading D2's sensor control and D3's manual recirculation have automated the damper? That is where a professional opinion earns its fee. It also tells you to build the provisional around F5, with variants and test data showing the heat saving.

Step 8: when to stop and how to record the search

You can stop when new strings and databases keep returning documents you have already seen, when citation chains loop back to your shortlist, and when you have covered patents and literature in every code you identified. Reaching page two is not a stopping rule.

Log as you go, not afterwards. A usable log has:

• the date of each search and the database used;
• the exact string, fields and filters, copied verbatim;
• the number of hits and how many you screened;
• each relevant document's number, publication date and the features it discloses;
• what you did not search, such as paid databases or non-English full text.

Dates matter because a search ages: applications still inside their 18-month window will surface later, so rerun the key strings before you file the complete specification.

Quality checks and common mistakes

• Searching only Indian records. InPASS is necessary but not enough; novelty is judged worldwide.
• Searching only in English. Classification codes, Espacenet's machine translation and PATENTSCOPE's cross-lingual search reach documents in other languages.
• Searching only titles. Titles are often deliberately bland; search abstracts, claims and full text.
• Stopping at the first 20 hits. Relevance ranking is not a novelty opinion, and Google Patents shows only one member of each simple family.
• Ignoring your own publications. First list every paper, poster, thesis, project report and demo video by your team, with dates.
• Confusing a prior art search with FTO. A prior art search asks whether your invention is new. A freedom to operate search asks whether making and selling your product infringes someone's in-force claims in a given country. One cannot answer the other.

A self-check: can you say which string and code found each shortlisted document? If not, rerun with a log.

How long it takes, and when to hand it to a professional

Budget a few focused days for a simple mechanical device, not an evening. Chemistry, biotech and software take longer because the vocabulary is looser and the literature larger.

Bring in a professional when the result decides real money, such as a PCT or foreign filing; when you have found close art and need a view on inventive step; when the terminology is messy; and when you need claims drafted. A good self-search makes that work cheaper and sharper, because the agent starts from your features, log and closest documents.

A self-search also improves a provisional filing. It tells you which feature to describe in depth, which variants and fallbacks to include, and what background to acknowledge. You then have twelve months to file the complete specification under section 9(1), time to rerun the search and refine the claims. Our guide on provisional vs complete specifications covers that choice.

How MYCrave can help

MYCrave Consultancy & Services helps where a self-search needs a second pair of eyes:

• Patentability searches as part of patent work: a review of your log and closest documents, further searching, and an opinion on novelty and inventive step before drafting;
• Provisional and complete specifications drafted around the features your search shows are different;
• Patent mapping studies for labs and R&D teams choosing a research direction;
• Freedom to operate searches when the question is whether you can sell a product, not whether you can patent it.

The examiner will run this search anyway

Every Indian application that goes to examination is searched. Your choice is when you see those documents: before you spend on drafting and filing, or first in an examination report.

So start small. Write the five features tonight, find two classification codes tomorrow, and keep a log from the first query. A dated, careful search is useful whatever it finds: a clear field tells you to file, and a crowded one shows you exactly where your invention differs.

Frequently asked questions

Keep reading