An AEO scorecard is a log of brand mentions and linked citations in a defined sample of AI answers. It can help you find incorrect facts and pages worth reviewing. It does not measure every buyer journey or prove that a brand is invisible. Keep observed answer data separate from a self-reported website checklist.
Review note, 22 October 2026: Unsupported audience-split, ranking-weight and turnaround claims have been removed. The worked numbers are illustrative, and checklist points are separated from observed answer measurements.
What’s changing about search?
Buyers can use conventional search, AI answers and other channels during the same journey. This guide has no evidence for a 50/50 split. AI answers can include links: Google documents supporting links in its AI features; ChatGPT search can show cited sources; and Perplexity describes answers with source links. A missing mention in one answer does not prove a lost sale or total absence from consideration.
Why do you need to measure AEO?
Measure a sample when it helps answer a business question: are important facts accurate, which pages are linked, and what changed after a correction? AI visibility tools already exist; this guide does not establish an easy competitive advantage or how many teams track the channel. Use this log alongside search, referral and conversion evidence, not as a replacement.
What is an AEO Scorecard?
Use the six fields below as a proposed review framework, not an industry-standard score. A monthly review can be a starting cadence. Record the date, exact prompt, engine and displayed mode or model, search setting, locale, account context and full answer or screenshot. Results can vary across runs; a small sample is not a market-wide estimate.
- Named or not. Record each brand named in the sampled answer and whether it was explicitly recommended. Keep branded prompts separate.
- Sample share of voice. Use a declared counting rule and denominator across your sampled answers, not an implied market-wide share.
- Cited sources. Save the actual linked pages and check whether each supports the answer. A citation does not reveal every source used.
- Accuracy. Compare facts, pricing and positioning with the current authoritative source; record the date and correction needed.
- Third-party mentions. Record relevant third-party pages and any observed citations without assuming a universal ranking weight.
- Change over time. Compare like-for-like samples, raw counts and repeated runs, while recording changes in test context.
The six-field manual log records observed answers. The linked questionnaire instead adds points from your own checklist responses. It does not query AI engines, inspect your website or establish whether an engine recommends your brand. Completing it is not a measured visibility baseline, and no completion time is guaranteed.
Open the self-checklist →How do you improve the score?
Start with accurate, accessible and useful content, then investigate the specific errors your log reveals. Google says established SEO practices remain relevant to its AI features and does not require special AI markup. No change guarantees a citation. Review third-party pages for factual errors without treating their mention count as a known ranking weight.
What does a first scorecard run look like?
Here is an illustrative spreadsheet example. The numbers demonstrate the arithmetic only; they are not results from a real brand or a statistically representative sample.
Step 1: pick the questions
Start with ten questions relevant to your buyers. Mix unbranded category and how-to questions with separately labeled brand comparisons. Branded questions can favor the named brands, so report them separately rather than presenting a combined rate as unprompted discovery. Use approved, non-confidential customer research for wording.
Step 2: run them and log each answer
For example, run ten questions once in each of two engines to collect twenty answers. Save the full responses and context. Record a brand mention separately from a linked citation and from an explicit recommendation. Where practical, repeat the same prompts in fresh conversations to see variation; do not treat one run as a stable ranking.
| Question | The exact wording you used |
| Engine | Which assistant answered |
| Named? | Yes or no |
| Competitors named | Every other brand in the answer |
| Sources shown | Any pages the engine links or cites |
| Accurate? | Are your price, features and positioning right? |
Step 3: do the math
In the illustrative sample, being named in five of twenty answers gives a 25% answer mention rate. For one defined share-of-voice calculation, count each distinct brand at most once per answer. If that produces forty brand-answer appearances across the sample and five are yours, your sample share is 5/40, or 12.5%. Keep the denominator and inclusion rule consistent. If there are no brand appearances, report share as not applicable. Compare like-for-like runs and show raw counts; a change alone does not prove your edits caused it.
Step 4: read the sources column
Read the linked sources and check whether they support the answer. Correct outdated facts on pages you control. On third-party pages, document any factual error and assess whether contact would be appropriate; absence from a list alone is not an error or an automatic outreach target. If no source is linked, record that limit rather than assuming you know which training or retrieval material was used. Our AEO vs SEO guide provides related review guidance.
The self-reported website checklist can organize your own answers about content and measurement practices. Its points are not an engine audit, a probability of citation or a substitute for the manual answer log above.
The takeaway
Use a defined answer sample to identify facts and source pages worth checking. Report what you actually observed, keep checklist points separate, and retain search and business outcome measures. A scorecard supports a review decision; it does not promise discovery, citations or sales.
Frequently Asked Questions
How often should I run it? A monthly review can be a starting cadence, with extra checks after an important factual change. Repeat prompts where practical to observe variation. This guide does not establish that monthly sampling removes noise or identifies a reliable trend.
Do I need a paid tool? You can start a small manual log using a spreadsheet and the access available to you. Check platform limits and any tool costs first. A paid product may help collect larger samples, but assess its coverage and method rather than treating its score as complete visibility.
What is an AEO scorecard? A proposed log of brand mentions, linked citations, factual accuracy, third-party sources and changes within a defined sample of AI answers. Record the exact prompts and testing context. A self-reported checklist is a separate planning aid, not measured answer visibility.
How is AEO different from SEO? SEO supports discovery in search, including linked AI features where applicable. This guide uses AEO to describe reviewing how a brand appears in AI answers. Answers can name multiple brands and include clickable sources; there is no universal single-winner rule.
Why bother measuring AEO if I already rank well? A search ranking and a mention in a particular AI answer measure different observations. Compare actual sampled answers, linked sources, referral traffic and available search reports. Do not infer complete invisibility or lack of measurement from one missing mention.
How long before AEO work shows up in the score? There is no supported fixed turnaround in this guide. Answers depend on the platform, query, context and available sources and may vary across runs. Record when a page changed and when you tested; a later citation does not establish causation.
Which questions should I test on an AEO scorecard? Choose a documented set of category, how-to and brand-comparison questions. Label branded prompts separately and keep wording consistent for comparisons. Ten questions is an illustrative starting sample, not a validated minimum.
How do I calculate AI share of voice? Define the counting rule before collecting answers. In this example, count each distinct brand once per answer, then divide your brand-answer appearances by all brand-answer appearances in the sample. Five out of forty is 12.5%. Keep raw counts and report no appearances as not applicable; this is sample share, not market share.
Get the templates behind these systems
The free starter kit has a template for each stage of the AI Content Engine, plus the audit to find your weakest link first. Everything else on the site is free to read, no email needed.