AI Visibility Vendor Test Script Template

AI Visibility Vendor Test Script (fill-in)

When to use: One row per vendor: identical prompt set loaded, engines actually shown, evidence export, quota fit and score, plus owners and a completion check.

AI visibility vendor test script

One row per vendor on trial. Every vendor gets the same prompt set, the same brands, the same location and the same test window. Only then are the scores comparable.

Buyer: [company] - Test owner: [name] - Prompt set version: [e.g. PS-v1, frozen on YYYY-MM-DD] - Test window: [start date] to [end date]

Per-vendor detail (fill for every row)

• Trial access: [free trial / paid month / demo account], started [date], ends [date]
• Prompt handling: [did the tool rewrite, merge or suggest prompts? list any change]
• Location and language set: [country / city + language]
• Engines claimed in writing: [list from the sales email or plan page + date]
• How answers are collected: [vendor's own words: API, interface, panel data, unknown]
• Run frequency on this plan: [daily / weekly / on demand]
• Spot-check result: [N answers re-run by hand, how many matched the tool's mention / no mention]
• Export file saved at: [folder or link]

Next action

• Test owner: [name] - task: freeze the prompt set and load it into every vendor on the same day by [date]
• Analyst: [name] - task: spot-check answers by hand and fill the coverage and export columns by [date]
• Budget owner: [name] - task: get quota terms in writing from each vendor by [date]
• Decision maker: [name + role] - task: approve one vendor or stop the purchase, with a reason, by [date]

Completion check

• Every vendor ran the same frozen prompt set version, and any tool edits are written down.
• Location, language, brand list and competitor list match across all vendors.
• Engine coverage lists what you saw in the tool, not only what the sales deck claims.
• Each vendor's export was opened, and it holds raw answer text, cited URLs and timestamps, or the gap is noted.
• Quota fit uses your real monthly volume and the vendor's written limits.
• The decision has a named approver, a date and a reason.

Identical Prompt Set Builder

When to use: Freeze one prompt set and one set of test settings so every vendor answers exactly the same questions under the same conditions.

Identical prompt set builder

The test is only fair if nothing varies but the vendor. Freeze the prompts, the brands and the settings before the first trial starts.

Frozen settings (same for every vendor)

• Prompt set version and freeze date: [PS-v1, YYYY-MM-DD]
• Your brand and spelling variants: [names, domains]
• Competitor list: [3 to 5 brands + domains]
• Country, city and language: [values]
• Test window: [start to end]

Rules

• Keep most prompts unbranded. Branded prompts only show how you are described.
• Paste prompts word for word. If a tool forces a rewrite, record it.
• Do not add vendor-suggested prompts to the scored set. Test them in a separate list.

Pull real buyer questions from the AI Buyer-Query Library Template, and check the wording people search with in Keyword Research.

Engine Coverage Check

When to use: Compare the engines each vendor claims with the engines that actually return answers in the tool, per plan and location.

Engine coverage check

A logo on a pricing page is not coverage. Mark an engine as covered only when you see stored answers for your prompts in the tool.

Questions to ask each vendor

• How do you collect answers for each engine? [answer + date]
• Which location and language are the answers from? [answer]
• Does the tool show which model or mode answered? [answer]
• Which engines cost extra on the plan we would buy? [answer]

Spot-check by hand

• Pick [N] prompts per engine and ask them yourself on the same day, from the same country.
• Answers vary between runs, so compare mention or no mention, not exact wording.
• Write down where the tool and your manual check disagree.

For a quick outside view of your domain, run the free AI Visibility Checker and keep the result next to each vendor's numbers.

Evidence Export and Quota Checklist

When to use: Check what each vendor lets you export and keep, and whether its prompt and refresh limits fit your real monthly volume.

Evidence export checklist

If you leave a vendor, the history should leave with you. Open every export file during the trial, not after you sign.

• Raw answer text per prompt, engine and run: [yes / no]
• Cited URLs and source domains per answer: [yes / no]
• Timestamp, engine and location on every row: [yes / no]
• Formats: [CSV / XLSX / API / PDF only]
• Screenshots or stored answer pages: [yes / no]
• History kept for: [period, from the vendor in writing]
• Export available on the plan you would buy: [yes / higher plan only]

Quota fit

Ask before you sign

• Does one prompt in four engines count as one credit or four?
• What happens at the limit: stop, overage or upgrade?
• Do unused credits roll over?
• Can we change prompts mid-month without losing history?

Evaluation Scorecard and Decision Record

When to use: Score vendors on the same criteria after pass or fail gates, then record the decision, the reason and the review date.

Evaluation scorecard

Gates first, scores second. A vendor that fails a gate is out, however good the dashboard looks.

Gates (pass / fail)

• Ran the full identical prompt set without forced rewrites.
• Shows stored answers for every engine you must track.
• Exports raw answers and cited URLs on the plan you would buy.
• Quota covers your monthly volume in writing.

Check claims against your own data

• Trackers measure AI answers. Traffic lives in your analytics.
• Google counts AI Overviews and AI Mode clicks in Search Console's Performance report under the Web search type. Its generative AI report shows impressions only, not clicks.
• GA4 has a default AI Assistant channel for chatbot referrals. Google says it excludes AI Overviews and AI Mode.
• Treat any vendor "AI traffic" number as an estimate until it matches these sources.

Decision record

• Chosen vendor and plan: [vendor + plan]
• Reason in one sentence: [reason]
• Approved by: [name + role] on [date]
• Re-test date: [date, rerun the same prompt set]

Track your own answers alongside the trial with the AI Visibility Repeated-Run Log. Paid seotools.pro plans include an AI Visibility Tracker; see pricing.

Spreadsheet Header (CSV)

When to use: A paste-ready header row and allowed values to run the vendor test in Google Sheets or Excel.

Paste-ready spreadsheet header

Paste this row into cell A1 of Google Sheets or Excel, then split text to columns by comma. One row per vendor, prompt and engine.

test_id,vendor,plan_tested,prompt_set_version,prompt_id,prompt_text,engine,location,language,run_date,tool_mention,tool_position,tool_cited_urls,manual_mention,spot_check_match,export_format,raw_text_exported,retention_written,quota_limit,quota_used,gate_result,score,owner,next_action,due_date,notes

Allowed values

• tool_mention / manual_mention: mentioned, cited, both, absent
• spot_check_match: match, mismatch, not_checked
• export_format: csv, xlsx, api, pdf_only, none
• raw_text_exported: yes, no
• gate_result: pass, fail

Sheet tips

• Filter spot_check_match to mismatch to see where a tool disagrees with what you saw.
• Pivot by vendor and engine to compare coverage in one view.
• Highlight rows where prompt_set_version differs. Those rows are not comparable.

Worked Example (fictional Quernwhistle field service software)

When to use: A filled two-week test of three invented trackers for an invented software brand, with one failed gate and one quota surprise.

Worked example: Quernwhistle

Example only. Quernwhistle is an invented scheduling app for HVAC contractors. Vendor A, B and C are invented trackers. All names, URLs, counts and dates are made up.

Buyer: quernwhistle.example - Test owner: marketing lead - Prompt set: PS-v1, 20 prompts, frozen 2026-10-26 - Window: 2026-10-27 to 2026-11-09 - Location: United States, English

What the test found

• Vendor B: AI Mode was on the sales slide but not in the tool. Raw answers needed a higher plan. It was dropped.
• Vendor C: the quota counted per engine, not per prompt. The plan they quoted would have run out mid-month.
• Spot-check: the analyst re-ran 10 prompts by hand. Vendor A matched on 9, Vendor C on 7.

Why this works

• The frozen prompt set made the merged prompts in Vendor B visible at once.
• Opening the exports during the trial caught the PDF-only limit before signing.
• Asking how credits are counted turned a vague quota into a real cost.

Want help setting up a fair vendor test, or a GEO program around it? Book a free consultation with Anatolii Ulitovskyi, or see our SEO and GEO services.