Setting a realistic AI visibility target
How to set an AI visibility target you can defend: start from your own baseline, use the leading competitor as a reference, set targets in prompts rather than percentages, and review them each quarter.
Our name has 99 in it, and we have never promised anyone a score of 99. Nobody can. AI engines decide what goes in their answers, change their models and retrieval without notice, and your competitors are working on the same prompts. The 99 is a direction.
But "no guarantees" is not a target either. Leadership still wants to know what success looks like by the end of the quarter, and a team without a target tends to celebrate any movement at all. Here is how we set a target that is ambitious and still defensible.
Why there is no industry benchmark to borrow
The first question is usually "what is a good score for our industry?" There is no honest answer to that, and we do not publish one.
A visibility score depends on the prompt set, the engines, the markets, the number of checks and the counting rules. Two companies in the same category with 30% each are measuring different things unless every one of those choices matches. Be wary of any average quoted without the method behind it. The only score you can compare yours with is your own, last month, on the same set.
Step 1: baseline for a full month
Run the agreed prompt set for a full month before setting anything. Record the score per engine, and sort the prompts into three groups: always named, sometimes named, never named.
Those groups are more useful for target setting than the percentage, because each one needs a different kind of goal.
Step 2: find the reference point in your own data
The leading competitor's score on your prompt set is a reasonable reference. It shows what the engines are currently willing to give a brand in your position, on these exact questions.
As an illustration: your ChatGPT score is 18%, and the strongest competitor on the same prompts scores 46%. Aiming for 46% in a quarter would be unrealistic. Aiming to close part of that gap is not.
Also count the prompts where no brand is named at all. Those set a ceiling. If 8 of your 50 prompts produce answers with no vendors in them, the maximum score on that set is well below 100%, and those prompts probably belong on a watch list rather than in the scoring set.
Step 3: set targets in prompts, then convert
Percentages invite vague goals. Prompts make the goal concrete. Using invented numbers on a 50-prompt set:
- Keep the 9 always-named prompts named
- Move 6 of the 14 sometimes-named prompts to named in most checks
- Get named at least occasionally on 5 of the 27 never-named prompts
Each line is something a team can plan work against. Then estimate what that would do to the percentage, and state that estimate as an outcome of the prompt goals, not as the goal itself. If the prompts move as planned and the engines change underneath, the prompt view still shows the work landed.
Step 4: set targets per engine
A brand at 30% in Perplexity and 6% in Gemini needs different targets for each. The work on a weak engine often starts further back, with crawler access or with being described correctly at all, so its first-quarter goal might be about access and accurate brand answers rather than the score.
Weight the engines by what they do for you. If your own referral data shows one engine sending most of your AI-referred visitors, its target matters most.
Step 5: tie every target to work
A target prompt with no page, citation plan or access fix attached is a hope. For each prompt in your target list, write down what will change: the page to be revised, the new page, the roundup to pursue, the crawler rule to fix. If you cannot name the action, take the prompt off the target list until you can.
Step 6: add leading indicators
The score lags the work. A page has to be fetched, then cited, then reflected in answers. So for the first month or two, set targets for the steps before the score: new pages fetched by AI crawlers, those pages appearing as cited sources, outdated cited pages corrected. These tell you the work is on track before the score can.
Step 7: review each quarter, and never move goalposts quietly
Review the target with at least two full months of data. Adjust if an engine changed significantly, if the competitor set changed, or if the prompt set had to change. Log every adjustment and the reason. A target that is lowered silently stops being a target.
What realistic sounds like
Realistic: "Get named at least sometimes on 5 of our 27 never-named prompts in ChatGPT and Perplexity by the end of the quarter, and keep the prompts we already hold."
Unrealistic: "Reach 70% visibility across all engines in 90 days." It ignores the baseline, the engines, the competition and the ceiling, and nobody can tell you what work would produce it.
How we set and track targets in Promptwatch
Targets for every client are set and tracked in Promptwatch. Prompt tracking produces the per-engine baseline and the three prompt groups. Competitive benchmarking and share of voice give the leading competitor's score on the same set, which is our reference point. Content gap analysis lists the prompts where competitors are named and the client is not, which is the starting list for the never-named targets.
We tag target prompts so they sit together in every view, and prompt trends show how each one moves between checks. For the leading indicators, Agent Analytics shows whether new pages have been crawled and traces the crawl-to-citation path, and citation analytics shows when they start being cited. None of that guarantees the number. It does mean that at the quarterly review, everyone can see which targets were hit, which were not, and why.