How to increase your share of voice in AI answers

What share of voice means when the answer names two brands

Ask ChatGPT for the best tool in your category and you get a short reply that names maybe two or three brands. There is no third page. That single constraint is what makes share of voice work the way it does in AI search. Across a whole set of category prompts, add up every brand mention, then look at what fraction belongs to you. That fraction is your share.

Here is the part that trips people up. Because each answer has room for so few names, one prompt is nearly zero-sum. Every time a competitor gets named, that’s a slot you didn’t get. So your share isn’t a score you improve in isolation. It moves when you take a mention that used to go to someone else. That reframes the whole job. You’re not trying to be good enough in the abstract. You’re trying to beat a named rival on a named question.

Start with the prompts you lose

Before you write anything, map where you stand. Run the questions your buyers actually ask a chatbot. Best [category] for [use case]. Alternatives to [competitor]. Cheapest option for a small team. Run each one a few times, because the answers shift between attempts, and note who gets named.

The prompts where a competitor shows up and you don’t are your target list. Not the ones you already win, and not some generic keyword report. This exact set of losing questions is the thing you work through. Sort it by how close the query sits to a buyer ready to choose, and start at the top.

Trace the mention back to its source

When a model names a competitor, something earned that. It might be a review site like G2, a Reddit thread that keeps getting cited, a “best tools” roundup from a publisher, or the competitor’s own page that answers the question cleanly. For each prompt on your list, work out which source drove the mention. On Perplexity you can often read the citations straight off the answer, which makes it the easiest place to see the wiring.

Once you know the source, the move is to earn comparable presence there. If a roundup names three rivals and not you, the work is getting into that roundup, or ones like it. If a Reddit thread is doing the lifting, the work is being a real, useful presence in those communities. Study the citations feeding the answers you lose, and you stop guessing about where to spend effort.

Publish content that answers those exact prompts

Your own site still matters, and this is where it pays off. Take the losing prompts and write pages that answer them directly, mapped to the real use cases behind them. Not a keyword. The actual question, phrased the way a buyer phrases it, answered in a self-contained passage a model can lift.

• Answer the question up top, in plain language.Models pull passages, not whole pages. If the answer to “which one handles X” sits three scrolls down, it won’t travel.
• Map each page to a use case, not a keyword.A page built for “best [category] for solo consultants” wins that prompt in a way a generic features page never will.
• Back the claim with something checkable. A number, a named customer, a source. Evidence-dense pages get quoted. Adjective-dense ones get skipped.

Fix the things that make you a risky pick

Sometimes you lose a prompt not because the content is missing but because your story doesn’t hold together. Your homepage says one thing, your G2 profile says another, an old press mention describes a product you no longer sell. When the sources a model reads disagree about what you are and who you’re for, you become the safe brand to leave out. It reaches for the rival whose description is consistent.

So clean up the entity. Make your category, your audience and your positioning match across the places the model looks. It’s unglamorous work, and it quietly decides a lot of close calls in your favor.

Re-measure on the same prompts

After you’ve earned a source, shipped the pages and tidied the entity, run the original prompt set again. Same questions, several attempts each. Did the share actually move? Did you start showing up on prompts you used to lose? That last check is the whole point. “I think it’s working” is not a measurement. A free AI visibility audit runs the same prompts across every major engine, so you can watch the number instead of guessing at it.

Set your expectations for the pace. Share of voice compounds slowly, and you win it prompt by prompt rather than all at once. Take one losing query, earn the mention, then the next. Do that across the list and your slice grows, answer by answer, at the expense of the brands who used to own those slots.