How to write a useful DeepSearch query

A useful DeepSearch query starts with the strongest identifier you actually know and puts it in the matching lookup mode. Use a complete name, exact email, full international phone number, exact username, company name or domain, valid VIN, or the original image. Add only a small amount of context that can separate real candidates. A long list of guesses usually makes the search harder to interpret, not more accurate.

Decide whether you need a lookup or a research question

The lookup field and Deep Search mode solve different problems.

• Use a lookup when you have a structured identifier and want candidates or a focused result: a name, username, phone, email, company, VIN, or image.
• Use Deep Search when you have a question that needs several public sources to be compared, such as a company's ownership history or what a public figure said about a topic over a stated period.

Do not paste an essay into a phone, email, username, or VIN field. Conversely, do not ask the research assistant to guess an identifier that you can enter exactly in its dedicated mode.

Use the most distinctive input you already possess

Choose the input that creates the fewest plausible matches.

| Starting information | Better query | Avoid | | --- | --- | --- | | Person's name | Full name plus one known public context signal | A first name or a biography made of guesses | | Username | Exact handle and, when useful, a platform focus | Assuming the same handle has one owner everywhere | | Phone | Full number with country calling code | A local-format number without a country | | Email | The complete address | Searching only a common local part such as info or alex | | Company | Legal or trading name plus jurisdiction or official domain | A generic industry phrase | | VIN | The exact 17-character VIN | A registration plate, owner name, or partial VIN | | Image | The clearest original file or direct image | A screenshot containing several unrelated images |

For a name search, useful public context includes a city, employer, school, profession, known handle, or official domain. Treat it as a disambiguation hint, not as a fact the result must be forced to match.

Format identifiers cleanly

Names

Use the person's full public name. If they use a stage name, former name, or a name in another script, run those as separate searches and compare the sources. Do not join several aliases into one oversized query.

Usernames

Enter the exact handle; the leading @ is optional. Platform focus can reduce noise, but it should not turn a handle collision into an identity claim. A cross-link from a known profile is stronger than visual similarity alone.

Phone numbers

Include the country calling code, such as +1, +44, or +971. A correctly normalized number lets DeepSearch distinguish the country and numbering plan. Carrier, region, and caller-label fields describe the line or a registry signal; they do not automatically prove who currently uses it.

Email addresses

Use the complete address exactly as received. An email search can find public connections and domain or incident metadata, but it cannot inspect the inbox. Never test a password-reset form to discover whether someone has an account.

Companies and VINs

For a company, add its jurisdiction or official domain when the name is generic. For a vehicle, enter the VIN as printed on the vehicle or its documents. A VIN lookup describes the vehicle; it does not identify its private owner.

Images

Use the least edited, highest-quality version available. Cropping, overlays, compression, and screenshots can remove useful context or produce weaker matches. Reverse-image search finds reuse and similarity; photo intelligence uses metadata and visual clues. Neither is a face-identification tool.

Write a bounded Deep Search question

A strong research prompt contains four parts:

1. Subject: the person, company, public topic, or VIN.
2. Question: the specific fact, comparison, or timeline you need.
3. Scope: a date range, country, product, event, or public role.
4. Evidence request: which claims need primary sources, dates, or direct quotations.

For example:

Compare the public statements made by Jordan Avery, the renewable-energy founder in Toronto, about battery recycling from January through June 2026. Separate direct statements from reporting and link each claim to its source.

The fictional occupation and city in this example are known disambiguators, not instructions to invent matching details.

Keep one subject and one purpose per search

Run separate searches when you have:

• two people with the same name
• an old and a new email address
• a personal and a company phone number
• several possible company entities
• an image collage rather than one target image

This makes the evidence trail legible and prevents a field from one subject from being read as if it belonged to another.

Interpret “nothing found” carefully

Before broadening a query, check for a typo, wrong country code, alternate name, or outdated identifier. Then try one justified variant. If the result remains empty, record that the available search found nothing; do not convert absence into a claim about the person.

The DeepSearch terms say results can be incomplete or inaccurate and must be verified before use. Continue with the common-name workflow or choose a different lookup type only when you have a real alternate anchor.