Good AEO targets come from your own baseline and your competitors' results on the same prompt set, not from industry averages. AI visibility numbers depend on which prompts you track, which platforms you test and how many times you run them, so a "benchmark" from someone else's method rarely transfers. This post shows how to build a baseline, measure the noise, benchmark against competitors, choose the right kinds of targets and set ambition levels you can defend.
Why universal benchmarks don't work
An AI share-of-voice figure is only meaningful alongside the prompt set that produced it. Change the prompts from "best CRM" to "best CRM for real estate teams" and every brand's numbers change. The same applies to platform mix, location, whether web search was on, and how many runs were averaged.
So treat published averages as context at most. The benchmarks that matter are:
- Your own baseline, measured with a fixed method
- Your competitors, measured with the same method on the same prompts
- Your own segments, for example category prompts versus comparison prompts, or one platform versus another
For how share of voice is calculated, see measuring share of voice in AI search.
Step 1: Build a baseline you can repeat
A baseline is only useful if you can run it again in exactly the same way next month. Lock down:
- The prompt set: 50 to 200 prompts, grouped by type and funnel stage. See how to build a prompt set.
- Platforms and modes: e.g. ChatGPT with search, Perplexity, Google AI Mode, Gemini
- Runs per prompt: at least three for important prompts
- Location and language
- Scoring rules: what counts as a mention, how position and accuracy are scored
Run the baseline over two or three weeks, not a single day, so one unusual day doesn't anchor your targets. If you need a quick first read before investing in tracking, start with a manual AI visibility audit.
Step 2: Measure the noise band
Before setting any target, know how much your numbers move when nothing changes. AI answers vary from run to run, so some movement is just noise. See why AI answers change every time.
A simple way to estimate it: treat each answer as a yes or no for "brand mentioned" and use the standard error of a proportion, √(p × (1 − p) ÷ n).
| Mention rate (p) | Answers (n) | Approximate 95% margin |
|---|---|---|
| 30% | 100 | ± 9 points |
| 30% | 400 | ± 4.5 points |
| 10% | 400 | ± 3 points |
Answers to the same prompt aren't fully independent, so treat these as the minimum noise. The practical rule: don't set a target smaller than your margin of error, because you won't be able to tell whether you hit it. Either set a bigger target or run more prompts and repetitions.
Step 3: Benchmark against competitors
Run the same prompt set for three to five competitors and compare like with like. This gives you the most useful benchmarks you'll have:
- Leader gap: your share of voice versus the category leader's
- Prompt-type gaps: where you trail most (often unbranded category prompts) and where you lead (often branded and comparison prompts)
- Platform gaps: strong in one assistant, weak in another, which usually points to a crawler, index or source issue
- Source gaps: domains that cite competitors but not you
See competitor benchmarking in AI search for the full method.
Step 4: Choose the right kinds of targets
Use three layers of targets: things you control directly, visibility outcomes and business outcomes. Mixing them stops a program from being judged only on the metric that moves slowest.
| Layer | Metric | Target form |
|---|---|---|
| Leading (you control) | Crawler access to priority pages | 100% of priority URLs fetchable by AI search crawlers |
| Accuracy on brand-fact prompts | No major errors across tracked platforms | |
| Coverage | Answer-first pages published for X priority topics | |
| Visibility | Mention rate on unbranded prompts | Up from baseline by more than the noise band |
| Share of voice vs leader | Gap reduced by a set amount | |
| Citation share for your pages | Rising on priority topics | |
| Business | AI referral sessions and conversions | Tracked trend with a growth goal |
| Self-reported AI attribution | Share of form responses naming AI tools | |
| AI-influenced pipeline | Reported as a range |
Different teams can own different layers; see AEO KPIs by team.
How ambitious should targets be?
Ambition should depend on where you start, because the work and the timeline are different in each case.
Largely absent from unbranded answers. Early targets should be leading ones: access fixed, errors corrected, key pages published, inclusion on the sources assistants cite. Visibility targets should be modest, such as appearing consistently on a defined subset of prompts.
Present but behind the leader. Target a reduction in the leader gap on your highest-value prompt group, with source-level goals (reviews, roundups, communities) that drive it.
Already the leader. Target defense: holding share of voice, accuracy near perfect, and expansion into adjacent prompt groups or platforms.
On timing, technical and accuracy fixes can affect search-grounded answers within weeks, while gains on competitive category prompts typically take months because they depend on third-party coverage. Set quarterly checkpoints rather than monthly pass-or-fail goals for those.
Example: a target sheet for one quarter
This example is illustrative; the numbers are made up.
A mid-market HR software company's baseline on 120 prompts, three runs each, across four assistants:
| Metric | Baseline | Noise band | Q1 target |
|---|---|---|---|
| Mention rate, unbranded prompts | 18% | ± 3 points | 24% or higher |
| Share of voice vs leader | 0.45× the leader | n/a | 0.55× the leader |
| Major factual errors on brand prompts | 6 | n/a | 0 |
| Priority pages fetchable by AI crawlers | 70% | n/a | 100% |
| Self-reported "AI assistant" on demo form | 4% of responses | small sample | Tracked, no target yet |
The unbranded target is twice the noise band, so a hit is real. The self-reported metric gets a baseline quarter before a target, because the form field is new.
Reviewing and resetting targets
Targets should be reviewed quarterly and reset when the measurement changes, not when results disappoint. Valid reasons to reset:
- The prompt set changed materially (new products, new markets)
- A platform launched or retired a major feature, changing what's measured
- You changed run method, location or scoring rules
When you reset, re-baseline and keep the old series for reference rather than splicing them together. Quarterly reviews are the natural home for this; see the AEO quarterly business review.
Mistakes to avoid
- Borrowing a vendor's "industry average" as your target
- Setting monthly targets smaller than the noise band
- Judging the program only on revenue in the first quarter
- Changing prompts mid-quarter and comparing against the old baseline
- Reporting a single run as if it were a measurement
The bottom line
Realistic AEO targets start with a repeatable baseline, an honest estimate of noise and competitor results measured the same way. Set leading targets you control, visibility targets larger than your margin of error, and business targets expressed as trends and ranges, then review them quarterly and reset only when the measurement itself changes. That gives leadership goals that are ambitious and checkable, which is what makes the case for continued investment; see making the business case for AEO.
