Practical Lens 73

AI recommends you.
Are you the right fit?

Your service is for companies only. Would AI recommend it to an individual buyer? Test whether the answer respects the conditions that decide who you can actually serve.

Published 29 October 2026 · By VerisAI

The short answer

Test one customer you can serve and one you cannot. Change only one important condition in the same question. Compare both answers with your current offer and the passages cited as evidence. Repair an identified information gap, then repeat the check. Being eligible and being recommended are separate observations.

1. Problem: a recommendation can be a poor fit

Illustrative example: a service accepts business customers only. An individual asks for the same work, and an AI answer recommends the company without checking that condition. The recommendation sounds useful, but the customer cannot buy that offer.

Business consequence: an unsuitable recommendation can create an enquiry that sales cannot fulfil and a customer expectation the company must correct. This example is hypothetical; no frequency or lost-revenue estimate is claimed.

What this lens checks

Lens 72 asks for evidence behind a visibility claim. This Lens tests whether a recommendation respects one known offer condition. It does not measure how often the company appears, rank competitors or infer that every omitted company was wrongly excluded.

Customer fit: The offer can meet the customer’s stated requirements under the company’s current conditions.
Material condition: A requirement that changes eligibility, such as customer type, supported market or an essential integration.
Not assessable: The answer does not assess the company or the condition, so it cannot establish whether the recommendation fits.

2. Verify: confirm the rule before testing

Open the current offer. Ask the commercial owner to confirm one condition that determines eligibility. Record the exact rule, a public source URL, the supporting passage and the date. If the rule is not public, record that gap; do not give the hidden rule to the AI test.
Create a pair. Keep the same company, service, market and wording. Change only the selected condition. One scenario meets it; the other does not.
Run both questions. Open an AI tool with web search enabled. Use separate fresh conversations without attached internal documents. Record tool, available model/settings, date and any personalisation you cannot disable.
Keep the evidence. Save complete answers and cited URLs. Open each relevant source and identify the exact passage. The commercial owner checks the conclusion against the confirmed offer.

Expected output: one verified offer condition, two saved answers and a fit assessment for each. Fresh conversations reduce carried-over conversation context; they do not establish identical or fully neutral conditions.

Two questions to copy

Replace the three bracketed fields with the same real company, service and market in both questions. This example tests a business-only offer. Use it only if that rule is true for the offer you are checking.

Scenario A: business customer

Would [company] be suitable for [service] in [market] for a business customer? Check current public sources, explain any important eligibility limits, and cite the passages supporting your assessment.

Scenario B: individual customer

Would [company] be suitable for [service] in [market] for an individual customer? Check current public sources, explain any important eligibility limits, and cite the passages supporting your assessment.

Open separate fresh conversations, select and copy each question, paste it, then submit. Do not include your expected answer or the private eligibility rule. For another condition, change that condition in both questions and keep everything else fixed.

3. Assess the result without overclaiming

How to label an observed answer
What you observe	Record	Next action
Recommends the offer despite a confirmed condition it cannot meet	Incorrect recommendation	Inspect the claim and its actual source
Rejects the offer using a demonstrably false condition	Incorrect exclusion	Locate and correct the misleading evidence where possible
Accurately recognises eligibility or the disqualifying condition	Consistent with offer	Retain the answer and source passage
Does not assess the company, omits the condition or has insufficient evidence	Not assessable	Do not count absence as an error; record what is missing

Important: an eligible company need not be recommended. A cautious answer asking for clarification can be appropriate. Judge the factual condition, not whether the answer promotes your brand.

4. Fix the demonstrated information gap

Missing condition? Put the real customer type, supported market or required integration beside the offer description. State important exclusions and exceptions plainly. Have the commercial owner approve the wording.
Conflicting pages? Align the service page, pricing information and FAQ with the same current rule. Remove stale claims. Add a clear link to the authoritative offer page.
External source is wrong? Save its exact URL and request a correction where appropriate. You control your own pages; an update there does not rewrite third-party content.
Evidence is inaccessible? Ask the web owner to verify the intended access to the public offer page and fix accidental restrictions. Keep private content protected.
Structured data disagrees? Correct existing relevant fields so they match the visible offer. Service audience and areaServed can describe intended customers and geography; they are not a guarantee of AI selection.

Assign an owner and retain the changed passage or setting. If the source is already clear and correct, do not invent a website defect to explain an incorrect answer. Keep the result as an observed answer error.

5. Retest: check the repair and the answer separately

After the change is live, reopen the affected pages and confirm the approved rule is visible and consistent. Then repeat both questions under the same available settings. Save full answers and sources on a second date as an initial repeat observation.

Repair verified: the documented content or access defect is resolved. Answer outcome: consistent with offer, incorrect recommendation, incorrect exclusion or not assessable. Record changed models, settings or sources that limit comparison.

Two scenarios are a diagnostic start, not a market estimate. Updated pages may not be retrieved immediately. A different answer does not prove that the repair caused the difference.

Evidence register

Use one row per scenario and run. Download the UTF-8 CSV, open it in your spreadsheet app, and fill the bracketed fields from saved evidence. Keep the answer capture and source passage available for review.

Download the customer-fit evidence register (CSV)

scenario,offer_rule,requirement,expected_fit,observed_result,platform,checked_at,source_url,source_passage,owner
A,[confirmed rule],[business customer],[eligible],[consistent/incorrect/not assessable],[tool],[date],[URL],[exact passage],[owner]
B,[same rule],[individual customer],[ineligible],[consistent/incorrect/not assessable],[tool],[date],[URL],[exact passage],[owner]

Optional Terminal check: missing evidence

Requires Python 3. Download the register, open Terminal in the same folder, paste the command and press Enter. With the untouched template, both rows should list the missing evidence fields.

python3 - <<'CHECK'
import csv
with open('lens-73-customer-fit-check.csv', encoding='utf-8-sig', newline='') as f:
    rows = list(csv.DictReader(f))
required = ['offer_rule','observed_result','platform','checked_at','source_url','source_passage','owner']
for row in rows:
    missing = [key for key in required if not row.get(key, '').strip() or row.get(key, '').strip().startswith('[')]
    print(row.get('scenario', '?') + ': ' + ('Needs evidence: ' + ', '.join(missing) if missing else 'Evidence fields filled; review accuracy manually'))
CHECK

Business meaning: this flags incomplete records. It does not validate the sources or judge customer fit. A completed row still needs a human review.

Optional PowerShell check: the same register

Open PowerShell in the folder containing the downloaded register, paste these commands and press Enter. With the untouched template, expect missing evidence fields for A and B.

$rows = Import-Csv -LiteralPath './lens-73-customer-fit-check.csv' -Encoding UTF8
$required = @('offer_rule','observed_result','platform','checked_at','source_url','source_passage','owner')
foreach ($row in $rows) {
  $missing = @($required | Where-Object {
    $value = [string]$row.$_
    [string]::IsNullOrWhiteSpace($value) -or $value.Trim().StartsWith('[')
  })
  if ($missing.Count) {
    '{0}: Needs evidence: {1}' -f $row.scenario, ($missing -join ', ')
  } else {
    '{0}: Evidence fields filled; review accuracy manually' -f $row.scenario
  }
}

This checks record completeness only. It does not query an AI service or establish that an answer is correct.

Sources and evidence boundary

Google Search Central: helpful, reliable, people-first content provides content-quality questions, including accuracy, clear sourcing and intended audience. It is Google Search guidance, not evidence that a specific AI tool will recommend a company.
Schema.org Service defines audience and areaServed fields for describing an offer. The audience property describes intended users. These definitions do not promise that an AI model will follow the fields.

Sources checked 29 October 2026. The paired-question procedure is an editorial diagnostic control. No live customer-fit experiment, error rate, lost-revenue estimate or causal effect is claimed.

About the publisher

VerisAI, founded in 2025, helps B2B services, SaaS, consulting and technology firms in Europe improve how their websites can be read, understood and cited by AI systems. Its AI Readiness Scoring & Monitoring service supports practical website checks. Contact support@verisai.eu.

Frequently asked questions

Does an AI recommendation prove customer fit?

No. Compare the recommendation with the current offer and its important conditions. A company can be named while a required market, customer type or integration is unsupported.

What should differ between the two questions?

Change one material requirement only, such as business versus individual buyer. Keep the company, service, market and question wording otherwise unchanged. Verify the actual offer rule before testing.

Does a missing recommendation mean a failure?

No. Eligibility does not guarantee a recommendation. If the company is not assessed, mark the result not assessable. Record an incorrect exclusion only when the answer rejects the offer using a demonstrably false condition.

What if the source is outside our control?

Save the incorrect statement and the exact source URL. Request a correction from its owner where appropriate and make your own current terms easy to find. Do not claim that editing your own page changes an external source.

Will clearer text or structured data fix every answer?

No. Clear, accurate public content reduces ambiguity in the source material. Structured data should match visible facts. Neither guarantees that an AI platform will retrieve it, use it or produce a correct recommendation.

Do two answers establish a market-wide problem?

No. Two scenarios provide an initial diagnostic check. Repeat observations with recorded conditions before describing a pattern. The exercise does not establish how many real customers saw an answer or prove that a website change caused a later difference.

AI recommends you.Are you the right fit?