AI Search Visibility Scorecard for Agency and Tool Selection

Audit Your Current AI Visibility

When to use: Establish a baseline of where and how AI engines currently mention or cite your brand before making changes.

Audit Your Current AI Visibility

You cannot improve what you have not measured. Before optimizing anything, find out whether AI answer engines currently surface your brand and for which questions.

• Brand: [Brand / domain]
• Engines checked: [Google AI Overviews, ChatGPT, Perplexity, Gemini, Copilot]
• Prompt set: [10–20 real questions your audience asks]

For each engine, run your prompt set and record the response verbatim.

1. Mentioned? [Yes / No]
2. Cited with a link? [Yes / No / source URL]
3. Accuracy of what it said: [Correct / outdated / wrong]
4. Who got cited instead: [Competitor / publisher]

Note honestly: AI answers vary by session, region, and model version, so treat any single run as a snapshot, not a guarantee. Repeat the audit on a schedule to spot trends.

Output: a prioritized list of high-value prompts where you are absent or misrepresented.

Make Content Quotable & Extractable

When to use: Restructure pages so AI engines can lift a clear, accurate answer directly from your content.

Make Content Quotable & Extractable

AI systems favor content that answers a question plainly and can be quoted without surrounding fluff. The goal is to make the correct answer easy to find and safe to lift.

• Target question: [The exact question this page answers]
• One-sentence answer: [Lead with this near the top]

Apply these structural patterns:

1. Direct answer first: state the conclusion, then explain the reasoning below it.
2. Clear definitions: define key terms in a self-contained sentence.
3. Scannable structure: descriptive H2/H3 headings, short paragraphs, and lists for steps or options.
4. Stats with sources: attribute every figure to a named, linked source and the date.
5. Plain language: avoid jargon that obscures the actual answer.

Reality check: being extractable does not guarantee a citation, but vague, padded content is rarely chosen. Write for a human who wants the answer fast.

Output: pages where the core answer survives being copied out of context.

Build Entity & Brand Authority

When to use: Make your brand a consistent, recognizable entity that AI systems can confidently identify and trust.

Build Entity & Brand Authority

AI engines map the world as entities (people, organizations, products) and prefer sources they can identify and trust. Consistency and credibility are the work here.

• Entity: [Brand / author name]
• Canonical description: [One consistent boilerplate used everywhere]

Strengthen the signals:

1. Consistency: identical name, description, and core facts across your site, profiles, and directories.
2. E-E-A-T: show real experience and expertise: named authors, bios, credentials, and a clear About page.
3. Be cited elsewhere: earn mentions on reputable third-party sites that AI systems already trust.
4. Authoritative references: ensure factual profiles on trusted reference and industry sources are accurate.

Honest framing: there is no switch that grants authority; it accrues from real reputation over time. Fix contradictions first: conflicting facts about your brand actively undermine trust.

Output: a coherent entity profile that AI systems can resolve without ambiguity.

Structured Data & Technical Foundations

When to use: Ensure your content is crawlable, clean, and clearly marked up so machines can parse it reliably.

Structured Data & Technical Foundations

AI engines and the crawlers feeding them need to access and understand your pages. Technical hygiene is table stakes, not a magic lever.

• Page / template: [URL or page type]
• Primary content type: [Article / Product / FAQ / How-to]

Cover the foundations:

1. Crawlability: confirm important pages are not blocked in robots.txt and review your stance on AI crawler access.
2. Clean HTML: render core content in HTML, not buried behind scripts that crawlers may skip.
3. Schema markup: add relevant structured data (Organization, Article, FAQ, Product, Breadcrumb) using valid syntax.
4. Semantics: one logical H1, ordered headings, descriptive link text, and meaningful alt text.
5. Performance: fast, stable pages that load without friction.

Reality check: schema helps machines understand context but does not force a citation. Validate your markup with Google's Rich Results Test or the Schema.org validator and fix errors before adding more.

Output: pages that are easy to crawl, parse, and categorize correctly.

Earn Third-Party Citations & Mentions

When to use: Build the external trust signals AI engines lean on by earning mentions in sources they already cite.

Earn Third-Party Citations & Mentions

AI answers frequently pull from sources the model already trusts: reputable publishers, reviews, communities, and reference sites. Your own pages are only part of the picture.

• Topic / query cluster: [Where you want to appear]
• Trusted sources in this space: [List sites AI already cites here]

Earn presence where it counts:

1. Find the sources: note which sites AI engines cite for your target prompts (from your audit).
2. Earn genuine mentions: pursue legitimate coverage, expert contributions, and honest reviews on those sources.
3. Manage reviews and profiles: keep listings, reviews, and community presence accurate and current.
4. Avoid manipulation: skip spammy link schemes. They risk reputation and rarely build durable trust.

Honest framing: you cannot control what third parties publish, and citations are never guaranteed. The realistic aim is to become a credible, recommendable option that trusted sources mention on their own merits.

Output: a presence in the sources AI engines already rely on for your topics.

Track & Measure AI Referrals

When to use: Set up the measurement you can realistically capture today and re-run your visibility audit over time.

Track & Measure AI Referrals

Attribution for AI search is still immature, so measure what you can and be candid about the gaps. Combine traffic data with repeated visibility checks.

• Analytics tool: [GA4 / your platform]
• Review cadence: [Monthly / quarterly]

Build a practical measurement loop:

1. Referral traffic: in GA4, filter the referral or session source/medium reports for AI hostnames such as chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com, and add UTM parameters to any links you control that AI tools might surface so those clicks are tagged.
2. Behavior: watch engagement and conversions from those visits, not just volume.
3. Visibility re-audit: rerun your prompt set on a schedule and log whether mentions or citations changed.
4. Brand and direct signals: note shifts in branded search and direct visits that AI exposure may influence.

Be honest: many AI interactions never click through, referrers are inconsistent, and no tool captures everything. Treat these numbers as directional. Document your method so trends stay comparable run to run.

Output: a repeatable scorecard of measurable AI visibility and referral signals.

AI Visibility Log (fill-in)

When to use: Record every question you ask an assistant: engine, model, location, date, run number, whether your brand was named, the citation URL and the evidence.

AI visibility log

One row per question you ask an assistant. Ask the same questions the same way every month, and the log becomes a baseline instead of a screenshot collection.

Columns

• Buyer query: [the question a buyer would actually type, not your keyword]
• Engine: [e.g. ChatGPT, Google AI Overviews, Perplexity, Gemini]
• Model or mode: [model name and whether web search was on]
• Location and language: [country, language]
• Date: [date asked]
• Repeat: [run 1, 2 or 3 of the same question that day]
• Brand mentioned: [yes / no] and where in the answer
• Citation URL: [the exact page cited, if any]
• Competitors named: [who else appeared]
• Evidence: [screenshot or saved answer link]

Log

Run it the same way every time

• Ask each question 3 times. Answers vary between runs, so a single run proves nothing.
• Use a signed-out or fresh session. Your own history skews what you see.
• Record the date and the model. Both change underneath you without notice.
• Save the evidence. A claim about last month's answer is unprovable without it.

What this cannot tell you

• Nobody can see an assistant's ranking factors. This measures outcomes, not causes.
• Answers are personalised and regional, so your log describes your sample, not the world.
• Being mentioned is not being cited. Track the citation URL separately, because that is the one that sends traffic.
• None of it converts directly to sessions. Treat AI visibility as a leading indicator, not a revenue line.

Agency & Tool Purchase Checklist

When to use: The questions to ask any vendor selling AI visibility, the answers to write down, and the signals that mean you should walk away.

Checklist for buying an agency or a tool

Use the log first, then use this. Take it to any vendor selling AI visibility and ask the questions in order.

Ask about measurement

• Which engines and models do you check, and how often? [answer]
• How many runs per question, and do you keep the raw answers? [answer]
• Do you record location and language? [answer]
• Can I see the evidence behind any number you report? [answer]
• Do you separate a mention from a citation? [answer]

Ask about the work

• What will you actually change on my site? [answer]
• Who implements it, you or my team? [answer]
• What is out of scope? [answer]
• What happens if the model changes and results drop? [answer]

Walk away signals

• Guaranteed placement in AI answers. Nobody can guarantee that.
• A ranking score with no method behind it.
• Numbers with no date, no model and no location.
• Refusal to hand over the raw answers they measured.

Decision

• Shortlist: [vendor or tool] · Scored: [score] · Reviewed by: [name]
• Decision: [buy / pilot / decline] · Date: [date]

Worked Example (baseline run)

When to use: Four filled-in rows from an invented baseline, including two runs of the same question that disagreed.

Worked example: one baseline run

Example only. The queries, answers and names below are invented to show how a row reads when it is filled in.

Read across, not down: we appear in 2 of 4 rows, and only 1 answer cited a page of ours. Run 1 and run 2 of the identical question disagreed, which is exactly why 3 runs is the minimum.

Baseline recorded: 3 September, and the same 12 questions get re-asked on the 3rd of each month.

Next step

Run a repeatable baseline with our free AI Visibility Checker, which asks several assistants the same buyer questions and reports where you were named. Then bring the gaps to a free consultation, or read how we do the work on the SEO and GEO services page.