How to Measure Share of Voice in ChatGPT (Free DIY Playbook)

Share of voice in ChatGPT is the percentage of relevant category prompts where your brand gets named in the answer. If you ask ChatGPT 50 questions a real buyer would ask in your category, and your brand shows up in 12 of those answers, your SOV is 24 percent. That is the whole definition. It is a category-level number, not a per-prompt rank.

It matters more than rank for one structural reason. Rank inside a single answer changes daily because LLM outputs are stochastic. SOV across a fixed prompt set changes monthly and tracks the underlying thing you actually care about, which is how often a buyer hears your name when they ask the model a question in your space. You do not need a $499 tracker to measure this. You need 50 prompts and a spreadsheet. This post is the playbook we use on our own book.

Build your prompt set

The first job is a fixed list of 50 prompts that a real buyer in your category would type. Fixed means you do not change the list once you start measuring. Drift in the prompt set kills the trend line. Split the 50 across five buckets so the answer set covers the full purchase path, not just the top of the funnel.

10 generic category prompts. These are the broad asks that surface the category leaders. Examples for a project management SaaS: best project management software for small teams, top project management tools 2026, popular alternatives to Trello, what software do marketing agencies use for project tracking, project management tools with good free tiers.

10 use-case prompts. These are the buyer-intent asks that filter the category to a specific job. Best project management tool for a remote five-person agency, project tracker for a freelance designer with three clients, tool that handles both project tracking and client billing, project software for a software team using Linear already, lightweight task tool for a solo founder.

10 comparison prompts. These force the model to name your competitors and rank them against each other. Asana versus ClickUp for a small marketing team, Monday versus Notion for project tracking, Linear versus Jira for a 10-person engineering team, Trello alternatives that are not Asana, what is the closest tool to Basecamp.

10 troubleshooting prompts. These catch the late-stage evaluator who already shortlisted but is hitting a wall. Why is Asana so slow for my team, how do I move from Trello to a real project tool, Monday is getting expensive what should I switch to, project tool that does not require a 50-seat minimum, alternative to Jira that non-engineers can use.

10 pricing and budget prompts. These are the bottom-of-funnel asks where pricing is the dealbreaker. Best free project management software, project tool under $10 per user per month, cheapest alternative to Monday, project software with a real free tier not just a trial, what is the most affordable team-collaboration tool.

Run the prompts

There are three ways to actually pull the answers, ordered by setup cost. Pick the one that matches the engineering muscle on your team.

Manual Monday morning. Open a fresh logged-out incognito window. Set the same region and the same model version every time. Paste each of the 50 prompts in order. Copy the answer into the sheet next to the prompt. This takes roughly 90 minutes once a month and costs zero dollars. It is also the least drift-prone option, because you control the session state directly.

No-code workflow. Make.com or Zapier can call the OpenAI API on a schedule, pass each prompt, and write the answer to a Google Sheet row. The setup is two scenarios and a webhook. Budget two hours to build, $20 per month in API spend at GPT-5 pricing for 50 prompts on a monthly cadence. The tradeoff is that the API response is not identical to the chat product answer, so this measures the underlying model output not the consumer surface.

Free Python script. Twenty lines using the OpenAI Python SDK plus gspread to write to Sheets. Same API tradeoff as the no-code path. Useful if you want to run multiple platforms (OpenAI, Anthropic, Google) from the same script. The Python route also lets you batch 200 prompts across four platforms in under five minutes once a month, which is what we use internally.

Whichever route you pick, run the full set once a month on the same day. The day matters because model versions roll on weekly cycles and you want the same point in the rollout each measurement.

Score the results

A mention is a literal brand name in the answer. Case-insensitive, plural and possessive count, partial matches do not. If the answer names ClickUp, that is one mention for ClickUp. If the answer says clickup-style tools without naming a specific product, that is zero mentions for everyone.

Ranking position inside the answer matters because the first-named brand carries more weight in the buyer's head than the fifth-named one. Three weights work for most categories.

First named gets weight 3. This is the brand the model pulled to the top of its candidate list. It is the one a buyer reading casually will remember.

Named in positions 2 to 4 gets weight 2. These are still on the shortlist a buyer will copy into a comparison spreadsheet.

Named in position 5 or later, or named only in passing, gets weight 1. These count as mentions but they are not driving consideration.

Score each of the 50 prompts for each brand in your competitor set. For a five-brand category that is 50 rows times 5 columns of weighted mentions. The sheet does the rest.

Calculate SOV

Worked example for a five-brand category. Your brand earns 18 weighted mentions across the 50 prompts. The four competitors earn 42, 35, 24, and 11 weighted mentions. Total weighted mentions across the category equal 130. Your SOV is 18 divided by 130, which is 13.8 percent.

Two reads matter from that number. The absolute level (is 13.8 percent high or low for a five-brand category) and the trend (is it rising or falling month over month). For a category with one runaway leader, the leader usually sits between 35 and 55 percent SOV. The second-place brand sits between 18 and 28 percent. Anything in single digits is a brand that needs distribution work before it gets a citation strategy.

Re-run monthly and track the diff

Same prompts, same model, same region, same day of the month. Build a second tab in the sheet that pulls the SOV for each brand each month and graphs it as a line chart. The interesting view is not the absolute number, it is the slope. A brand that moves from 8 percent to 14 percent in three months has shipped something that is working. A brand that drops from 22 percent to 17 percent in the same window has a problem that does not show up anywhere else in the funnel yet.

Annotate the chart with what happened each month. A Reddit AMA, a paid ChatGPT Ads test, a new pricing page, a new comparison article. The annotations are how you tie a marketing input to an SOV output. Six months in, the chart is the single most useful artifact in the marketing reporting stack, because it is the only metric that survived the death of search referrer data.

Free Google Sheets template

We built the template we use internally and made it free to copy. It has the five prompt buckets pre-labeled, the weighted scoring formula wired up, the SOV calculation across a competitor set of up to eight brands, and the monthly trend chart. Add your prompts, paste the answers, the sheet does the math.

Grab it from the AI visibility checker. Same page also runs a free single-prompt check against your brand if you want a quick read before you build out the full 50-prompt set.

When DIY breaks down

Three failure modes push teams off the spreadsheet and onto a paid scanner.

Platform sprawl. Once you are tracking ChatGPT, Perplexity, Gemini, AI Overviews, and Claude on the same prompt set, the manual rerun is a full day of work. The math is the same, the legwork is not. A paid scanner that hits four platforms in 15 minutes is worth the money at that point.

Reporting cadence. A monthly measurement is fine for an internal trend line. If you are reporting to a client or a board weekly, the spreadsheet falls behind. Daily refresh is impossible by hand.

Citation source attribution. The spreadsheet tells you whether you got mentioned. It does not tell you which third-party source the model pulled the mention from. Profound, Peec, and Scrunch will name the source URL, which is the input you need to plan the next round of seeding work. If your strategy is source-driven, you need that data and the spreadsheet cannot produce it.

Below those three thresholds, the spreadsheet is the right tool. Above them, the $499 per month tracker pays for itself in operator time. Most teams we work with cross the threshold around month four, once the trend line has been worth reading for a quarter.

How paid placement moves SOV faster

Every tracker we just talked about, paid or DIY, measures the same surface. The number moves when one of two things changes. Either the model retrains and pulls in new third-party sources where you are mentioned (organic, slow, compounds for years). Or paid placement puts your brand into the answer directly (paid, fast, stops when the budget stops).

The paid lever is the one most teams underweight in the SOV conversation, because the trackers themselves only measure organic citation. ChatGPT Ads went live for the US, Canada, Australia, and New Zealand earlier in 2026, with the UK, Mexico, Brazil, Japan, and South Korea announced on May 7 for rollout in the coming weeks. The EU, India, MENA, and most of APAC are still gated for regulatory reasons (GDPR, AI Act, DSA). Ads only render to Free and Go tier users. Pricing is cost-per-engagement with semantic intent targeting, not keyword bidding. OpenAI's direct minimum is reported at $200K, which is why managed access is the actual route for most operators.

At one Ranqer client we are seeing roughly 50 percent of incoming ChatGPT-ad clicks convert to paid customers in their Google Analytics. That figure is a single-client internal observation, not a benchmark across the book. The point is not the 50 percent. The point is that the channel converts differently from Google or Meta because the user arrives mid-intent from inside an answer, not from a feed or a SERP.

If your SOV chart has been flat for two quarters and your organic citation pipeline is already running, paid is the lever that moves the number this month instead of next year.

Frequently asked questions