How to Use AI in a Small Business: Perplexity for Research Jobs

To use Perplexity for market and competitor research, write a short brief that names the decision, the questions and the kinds of source you accept, run it, and then open every cited page before keeping any figure. Perplexity finds and summarises pages quickly; it does not confirm that the summary matches the page, so the checking is your job.

Prompt lists make this look easier than it is. What separates useful research from a confident-sounding paragraph is the brief you give it and the checks you run afterwards. The walkthrough below uses one real-feeling decision, a bicycle repair shop considering e-bike servicing, to show the brief, a sample answer, the source checks, and how to turn the result into a price and a supplier shortlist.

Follow me on Instagram@sagnikteaches

The decision that frames the research

A three-person bicycle repair shop gets a few e-bike enquiries a month and turns most of them away, because the staff have not been trained on motor systems and the shop does not own a diagnostic tool. The owner wants to know whether to offer e-bike servicing, at what price, and from whom to buy the equipment. That is one decision with three research jobs: competitors, pricing and suppliers. Every figure that follows is illustrative.

Connect on LinkedInSagnik Bhattacharya

Start by writing the decision in one sentence: "Should we add e-bike servicing within three months, and if so at what price and with what equipment?" Everything you ask Perplexity should serve that sentence. A question that does not, however interesting, is drift.

Subscribe on YouTube@sagnikteaches

Writing the brief before you open the tool

A brief has five parts. Write them in a note, then paste them as the first message:

1. The decision, in a sentence.
2. The questions, numbered, answerable from public pages. "Which independent shops in [my town and the next two] advertise e-bike servicing, and what do they say they cover?" is answerable. "Is e-bike servicing a good idea?" is not.
3. What counts as a source: a business's own site, a manufacturer's page, a trade body, a published price list, a recent news report. State what does not: undated forum posts, affiliate "best of" lists, anonymous review sites.
4. The output format: a table with the claim, the source address, the date of the page and a confidence note.
5. The rule for gaps: where nothing is found, say "not found". No estimates.

That fifth item is the one most people leave out. Without it, a model will fill an empty cell with something plausible, and a plausible number in a table is the most dangerous thing in this exercise.

The brief as a prompt

I run a three-person bicycle repair shop in [my town]. Decision: should we add e-bike servicing within three months, and at what price?

Research these questions using public web pages only:
1. Which shops within about 30 minutes of [my town] advertise e-bike servicing? For each, what services and prices do they publish?
2. What do manufacturers of common e-bike motor systems say about who may service them (authorised training, tools, warranty conditions)?
3. What diagnostic tools do those manufacturers or shops mention?

Accept as sources: business websites, manufacturer pages, trade bodies. Do not use undated forum posts or affiliate lists.
Output a table: claim | exact page address | page date or "undated" | what the page actually says (quote under 25 words) | confidence (high if the page states it plainly, low if inferred).
If something is not found, write "not found". Do not estimate.

The "quote under 25 words" column forces the model to show evidence you can compare with the page. It also makes made-up claims easier to spot, since a fabricated quote usually fails a quick search of the page.

Reading a sample answer critically

Here is the shape of an answer you might get, with invented placeholders in place of real businesses:

Read it as a sceptic. The table has five rows, and each needs a verdict before any of it enters your decision. Row one needs an open and a date check. Row two helps, but it tells you only that one shop is certified for one brand. Row three is the most useful row for the decision, since it suggests the manufacturer controls who can do motor work, and it deserves the closest read. Row four comes from a directory, which may be stale. Row five is the one to suspect: a tool price with no source is a guess, and the model's own table says "not found" in the page column while still offering a figure. Discard the figure.

Checking sources: a five-minute routine for every row

Open each address in a new tab. For each, answer five questions. Most claims fail at the first or the third.

1. Does the page exist and load? Dead or redirected links show a fabricated or moved citation.
2. Is the claim on the page? Search the page for the key words. The quoted words should appear, or the summary should be a fair reading.
3. How old is it? A price list with no date, or a footer saying 2022, is stale for pricing. Prices change, and a competitor's page often lags its real rates.
4. Who wrote it? A business describing itself is a fair source for what it offers, a poor one for how good it is. A directory copying someone else's page is a weak source.
5. Could it be a promotional claim? "Leading specialists" is marketing; "from $65" is a price.

Mark each row kept, kept with a note, or dropped. In the example, rows one and three survive, row two is kept with a note ("one brand only"), and rows four and five go. Three firm facts from five, which is a normal yield and a good reason to ask for more questions than you need. The method is set out in more detail in how to check sources and citations in AI research.

The failure that looks like success

The nastiest case is a real page that does not say what the answer says it does. A citation to a manufacturer's dealer-finder page, offered as proof that "dealers are required for repairs", sends you to a page that lists dealers and says nothing about repairs. The link works, the page is official, and the claim is unsupported. The only defence is reading what the page says. If you are in a hurry, check the claims that affect money first.

Pricing research: turn found prices into a range, not a guess

After the competitor table is checked, you have a handful of published prices. Put them in a sheet with columns for shop, service, price as published, whether it is "from" a price, the date of the page, and the address. Illustrative data for a full e-bike service: $110 ("from"), $120, $135, $150 ("from"), and $90 from a page dated three years ago. Drop the old one, since it is not current evidence. The four current figures give a range of $110 to $150 and a middle near $127.

Two cautions. A "from" price is a floor, not what customers pay. And a published price tells you what competitors ask, not what they charge after extras, so note which prices include parts and which do not. The comparison is only fair when like meets like. The pricing question itself, what to charge, is yours; what AI pricing research can and cannot tell you sets out why. Keep the checked sheet and refresh it monthly rather than starting again each time.

A short sum for the decision: if a full service takes 1.5 hours of a mechanic's time costing $22 an hour all in, plus $12 of consumables, the cost to deliver is $45. At $127 the gross margin is $82 a job, which at four jobs a week is $1,312 a month before the equipment. If the diagnostic tool is $1,200 and training is $400, the shop recovers both in a little over a month on those numbers. The sum depends on the four jobs a week, which is a guess you should test, perhaps by counting how many e-bike enquiries came in over the last quarter.

Supplier research: use Perplexity to find names, then verify elsewhere

For suppliers, Perplexity is a decent way to build a first list, and a poor way to confirm any of them. Ask for named companies that sell the kind of tool or part, with a link to each company's own site. Then do the following for each:

• open the company's own site, not the citation, and confirm that it sells what the answer claimed;
• find a company address, a way to contact them and terms of sale, since a shop front with none of these needs caution;
• look for an independent mention, such as a trade forum thread or a manufacturer's partner list, that is not written by the supplier;
• request a quote rather than relying on a listed price.

Ask the tool for what it could not find: "Which suppliers did you consider and reject, and why?" The answer sometimes reveals that a result came from a marketplace listing, not a dealer. Quotes then go through the routine in comparing supplier quotes line by line, so that price, minimum order, freight and terms line up.

One privacy point belongs here. Keep your own pricing, margins and quotes out of prompts to a consumer tool. Perplexity's "AI Data Retention" setting is on by default on its Free, Pro and Max plans; it sits under Preferences in your account settings, and switching it off applies only to data collected from then on, so decide before you start, not after. Public-page research does not need private data.

Keeping a research project, and what the plans give you

Once a question recurs, store it. Perplexity's help centre describes a feature called Projects (older guides call it Spaces; check the name in your own account): you add instructions of up to 8,000 characters and upload files, with up to 10,000 files and 5 GB per project and 100 MB per file, and projects are private by default. That is enough to keep the brief, your source rules, and last month's checked table in one place, so each monthly refresh starts from the same standards.

On plans: the Pro page showed $17 a month, billed annually, with no monthly price displayed, when checked in November 2026. A free plan exists, and the higher Max tier showed $167 a month billed annually on the same page. Names and limits change, so confirm them on the pricing page before choosing, and test the free version with your brief first. For help deciding whether the paid plan earns its price, read whether Perplexity Pro is worth paying for in a small business, and for a head-to-head, Perplexity versus ChatGPT for business research.

The second pass: ask the tool to attack its own answer

Once the first table is checked, run a second prompt that asks for the opposite view. It is cheap, takes two minutes, and finds gaps the first pass hid. For the bicycle shop:

Using only public pages, find evidence AGAINST the idea that a three-person bicycle shop should add e-bike servicing: restrictions from manufacturers, warranty risks, insurance or liability issues mentioned by trade bodies, and shops that stopped offering it. Same table format. If you find nothing, say "not found" rather than reasoning from general knowledge.

Illustrative return: two rows, one from a manufacturer's dealer-terms page saying that unauthorised work on the battery pack may affect warranty, and one from a trade body's page on handling lithium batteries safely in a workshop. Both are worth reading in full. Neither changes the plan to service brakes, drivetrain and tyres, but together they shape what the shop should not do: open a battery pack. Without the second pass, the owner would have seen a one-sided table full of price points.

A third pass, for numbers: ask which of your checked claims are about prices or dates and which about capabilities, and list the five facts that would change your decision if they were wrong. Then check those five twice. It sounds fussy; it is the nearest thing to quality control a one-person research process has.

How long it takes and how often to repeat it

For a job of this size, expect about 15 minutes to write the brief the first time, 5 minutes to run it, and 45 to 60 minutes to check sources. The checking dominates, as it should. The second month is faster, because the brief and the source rules already sit in a project and only the dates and changed prices need attention. A sensible rhythm for prices is monthly; for suppliers, twice a year or whenever a supplier raises its prices; for competitor services, quarterly. Put the dates in a calendar, because a research habit that depends on remembering is not a habit.

A mistake that went into a price list

An illustrative case: a solar installer researching battery costs asked Perplexity for typical installed prices for home batteries in its area. The answer cited an installer's news page, gave a confident range, and the owner quoted the low end to a customer. The page, once opened, turned out to be eighteen months old and described a subsidised price that no longer applied. The quote was honoured at a loss on that job and the owner fixed the process: no price enters a quote unless the source page is under six months old and states the price on the page itself, in a sheet with the date beside it.

The same installer now runs a monthly refresh. The brief lives in a project, the sheet keeps the last three months of figures, and any change of more than 10% since the previous month is flagged for a human look. The point of that habit is that the research becomes a record, so a surprise shows up as a change, not as a mystery.

Research jobs in other businesses

The method carries to other trades, with different questions and different dangers.

A bakery thinking of adding a wholesale line to cafes asks: which cafes in the area list a local baker as a supplier, and what do their menus show for pastries? Menus are a primary source, and prices on them are current. The risk is stale social posts, so the brief says "pages from the last twelve months only".

An HVAC contractor weighing a new heat-pump brand asks about installer requirements, warranty terms and published retail ranges. Manufacturer pages are the key sources, and the check is the date, since specifications change between model years. Here the tool is best used to find the manufacturer's own documents, which you then read yourself rather than accept a summary.

A small online shop researching a rival's range uses the rival's public product pages, noting count, price points and shipping thresholds. The question to avoid is "how much do they sell?", which no public page can answer. Estimates from third-party traffic tools are guesses, and the table should say so.

What to do with the findings

End every research job with a one-page summary written by you, from the checked table only: what is established, what is uncertain, and what you would do next. Ask the assistant to tidy your draft if you like, but give it the checked table and not the original answer, so unchecked claims cannot return by the back door. File the summary, the table and the brief together, with the date.

For the bicycle shop, the one-pager reads: four current published prices for a full e-bike service run from $110 to $150; one brand's manufacturer restricts motor repairs to authorised dealers, so the shop would be limited to non-motor work on that brand; a diagnostic tool and training would cost roughly $1,600 by the owner's own quotes, to be confirmed; decision: offer a "brakes, drivetrain and tyres" e-bike service now at $95, and ask two manufacturers about dealer status. That decision came out of checked facts and the owner's judgement. The AI saved an afternoon of searching, and nothing in the one-pager is unsourced.

If you want a research routine like this set up around the decisions your business makes every month, that is what my AI implementation consultation covers, and you can find it at the contact page. A broader version of the same idea, across more kinds of research, is competitor research with AI in one afternoon.

Perplexity questions small business owners ask

Is Perplexity better than ChatGPT for research?

For questions where you need linked sources on the page, Perplexity's answer format is convenient, because each claim carries a citation. ChatGPT and Claude also search the web and are stronger at working with your own files. Use whichever shows its sources clearly, and open them either way.

Do I need a paid plan for competitor research?

Not to start. Run the same brief on the free tier and compare the depth of the answer with a paid plan before committing. Perplexity's Pro price page showed $17 a month billed annually when last checked, with no monthly price visible, so confirm what you would pay on the current page.

Can I paste supplier prices or customer details into it?

Avoid it. Perplexity has a setting called AI Data Retention that is on by default on its Free, Pro and Max plans; you can switch it off under Preferences in your account settings, but the opt-out only covers data collected afterwards. Research on public pages needs no private data, so keep confidential quotes and customer details out of the prompt.

How do I know a citation is real?

Open it. Check that the page exists, that it is dated, and that the sentence you were given appears there or is a fair summary of it. Citations can point to a page that only loosely supports the claim, so a link alone is not proof.

Further reads

• How to Test a New Service Idea With AI Before You Launch It — Test a new service idea before you commit money.
• AI Supplier Management: Track Prices, Lead Times, and Risk — Track supplier prices and lead times once chosen.
• Hotel Rate Shopping: A Weekly Competitor Rate Sheet Built With AI — A weekly competitor rate sheet, built step by step.
• How to Do Keyword Research With AI for a Small Business Website — Research what customers search for, with the same checks.
• How to Find B2B Prospects With AI Research Tools — Use the same research habits to find buyers.
• How Consultants Use AI for Client Research Before Discovery Calls — Research a client or buyer before a first call.
• How to Track Competitors' Prices and Offers With AI — Keep the checked price sheet current each month.
• Best AI for Consulting Research: Market Scans With Source Checks — A source-check routine for consultants: prompt chain, source log, three verification tiers and how ChatGPT, Claude and Gemini research modes differ.
• Best AI for Content Research: Briefs and Source Checks for Agencies — Compare Perplexity, ChatGPT, Claude and Gemini Notebook for agency research, then add a source-check step that stops unverified claims reaching a client draft.
• ChatGPT for Sales and Marketing: B2B Battle Cards and One-Pagers — Prompts and a fact-check routine for a competitor battle card and a one-page service sheet in a small B2B firm.
• Best AI for Startup Research: Customers, Rivals and Pricing — A comparison of search-enabled assistants for startup research, with a competitor table and a routine for checking every claim at its source.
• Competitor Battlecards for a Small B2B Team: AI-Drafted, Rep-Checked — A competitor battlecard process a five-person sales team can run: AI drafts from public sources, reps verify, a monthly trigger list keeps it current.
• Perplexity Merchant Program: Should a Small Online Store Join? — What Perplexity and feed-tool guides say about the free programme, who can apply, what a product feed needs, and a sum to decide if the prep time pays.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Perplexity help centre (Projects article; Data collection at Perplexity) and Perplexity pricing page, checked 5 November 2026.