How to research an ideal customer profile from public evidence

Start ICP research with one offer, one candidate segment and one decision. Write what the offer helps customers do, what they need in place to use it and what would make the engagement a poor fit. If you plan to use the profile for lead qualification, name the facts you could check in a discovery call or account record.

For a service that turns web evidence into client briefs, a team with recurring research and source requirements is a candidate segment. Frequency matters because it tells you whether the workflow will be used again. A buyer persona addresses a different level of the decision. It describes a person involved in the purchase, their responsibilities and the proof they need. Keep the ICP focused on the account and its situation.

Choose each public source for a question it can answer. Company service pages show what a business says it offers. Reviews and public discussions can describe why somebody tried a tool or struggled with it. Search results help find those sources. Open the original page when you need the complete wording or context. Search results versus page extraction

Set the scope before gathering records. Note the products or accounts you selected, the dates, the source types and any sort order you can see. Include favorable and critical experiences. A positive review can explain what made a workflow usable. A critical one can reveal conditions a tool needs to meet. Keep repeat posts or syndicated copies together so one experience does not appear to be several.

With MCP Scraper, use Google search for discovery, page extraction for a selected public URL and Google Maps collection for reviews attached to a specific listing. Check that you opened the intended listing or URL and that the results contain the fields you need. Record failed or missing sources in the same ledger as the returned records. Google Maps scraper

A useful ICP research ledger keeps each observation traceable. Give the row an evidence ID. Save the exact source link, what you requested, what came back, when you captured it and which date the source displayed. Write a short observation in one field and any possible interpretation in another. Mark fields you cannot verify as unknown.

A published MCP Scraper example requested up to five review cards from the Formula Roofing and Remodeling listing in Denver. It returned five. One card showed a five-star rating and a relative date of “5 months ago” at the time of capture. The public sample records the exact listing, capture time and requested and returned counts. Keep the relative date as displayed unless you can verify the exact review date. Download the captured example

Formula Roofing lists residential and commercial roofing. The review gives you a positive rating and leaves buyer type and purchase trigger unknown. If you have permission, check the matching estimate intake or ask the customer what kind of property it was and what prompted the request. If the record identifies a commercial property, keep that observation out of the homeowner sample. Leave either field unknown when you can’t verify it.

An ICP qualification signal is an observable fact that helps you decide whether an account may fit an offer. To propose one, ask what job the observation reveals, why that job matters to fit and how a person on your team could check it.

In MCP Scraper’s category research, an Apify reviewer described using scraped material for paid-media audience research and reported duplicates and missing field documentation in a September 29, 2026 Trustpilot review. Another Apify reviewer described an AI-linked workflow moving to a more expensive route without an upfront check in an October 5, 2026 Trustpilot review. MCP Scraper captured both Trustpilot reviews on November 7, 2026. The individual links did not open in our November 10 check. See the November 7 capture notes for a paraphrase of each report and its original link. A Firecrawl reviewer wanted to look up attributes missing from an initial extraction in a December 23, 2025 Product Hunt review by Pranav Pai Vernekar. These are individual reported experiences. Together they suggest questions worth testing about recurring research, evidence quality and cost visibility.

For a scraper that helps teams assemble briefs from public sources, validate agency teams that repeat client research before teams that only need a one-off lookup. The public reviews point to work involving several records or missing fields. They give you a reason to test the recurring workflow first. They leave the reviewers’ account fit, budget and willingness to buy unknown.

Reverse that order if the agency already gets the source evidence it needs from its current process and the other team shows a recurring gap. Check the workflow in a recent deliverable before treating either account as a fit.

Keep the denominator with any count. If three of ten selected reviews mention missing fields, write “three of ten reviewed records.” Public reviews reflect who chose to post and what you selected. Market size, willingness to pay and typical outcomes need their own evidence.

A working ICP gets stronger when its qualification signals agree with evidence about accounts you know. Where you have permission, compare the proposed signals with customer interviews, closed-won and lost deals, repeat usage and retention. Look at accounts that had the proposed signal and still struggled. Check whether a different factor, such as access to the required source or an internal approval step, explained the outcome.

Write a decision for each signal. Keep it if it helps explain fit across the evidence you have. Narrow it if it only applies to a particular workflow. Leave it open if you have no way to check it yet. A marketer with no CRM access can hand over a working profile and a short list of interview questions for the client to validate. Further reading on ICP validation

Record account fit and buying timing separately. If the agency confirms that client briefs regularly need checked source links, mark it as a candidate fit. If it has no upcoming brief or specific collection problem to test, leave buying timing unknown. Recheck when the agency has a dated brief or a collection problem that a scraper pilot can address.

For a first scraper pilot, use one public source from the upcoming brief. Agree on the fields a reviewer needs, record any field the scraper leaves blank, and ask a teammate to reopen the original source before using a finding. If the source or required fields cannot be checked, keep the account fit provisional and test the missing prerequisite. Choose a first scraper test

Save the ICP and its evidence ledger as separate records. The ICP should tell the team which accounts to prioritize and how to check the criteria. The ledger should let another researcher reopen the source, see what it said and decide whether the interpretation holds. Record what changed when new interviews or account data revise the profile.

If you use an AI assistant to group observations, ask it to return the evidence IDs for each proposed theme, list conflicting examples and leave unsupported fields blank. Read the underlying sources before adopting a qualification rule. When another research session will need the material, you can deliberately retain the selected sources in Memory Vault and verify retrieval with the next research task. Saving scraped research in Memory Vault