AI Share of Voice or Organic Keyword Rankings: Which Reads Demand?

Neither one reads demand on its own. Plain keyword rankings count where you sit for a list of keywords you chose, and Semrush notes that such counts treat a rarely searched keyword the same as a heavily searched one. AI share of voice, as defined here, counts how often your brand appears in AI answers to a set of prompts you chose and gives every prompt the same weight. AI engines do not publish how often each prompt is asked; some vendors estimate prompt interest, but those figures are modelled, not engine-published. Demand comes from query data. Both signals read what happens on the questions you pick, so the useful move is to pick the questions from demand evidence and then read both.

Key Takeaways

• Rankings and AI share of voice are outcome readings, not demand readings. Each measures your position on a question list. As defined here, neither measures how many buyers ask.
• In Semrush's share-of-voice definitions, only the SEO version carries a demand weight; its AI share of voice is a share of mentions. The SEO version estimates organic traffic from rankings, search volume, SERP features and expected click-through rates.
• The two result sets overlap only in part. Three studies below measure that overlap on different engines and dates; none shows that ranking causes or fails to cause citation.
• Do not retire either signal. A ranking is one read of whether a search engine can retrieve your page for the question. AI share of voice is the read of what the answer then says.
• Choose and weight questions from query data, such as your own Search Console queries, sales-call questions and support tickets, before measuring either signal.

What does each signal actually count?

Each one counts presence on a list you wrote, and, of the three signals as defined here, only SEO share of voice adds a demand weight. The differences decide what each can and cannot tell you about intent.

Signal	What it counts	Weighted by demand?	What it cannot tell you
Plain organic keyword rankings	Your position for each tracked keyword	No. A rarely searched keyword counts like a heavily searched one	Which rankings matter to revenue, or what buyers ask in full sentences
SEO share of voice	Your estimated share of organic traffic across tracked keywords, from position, search volume, SERP features and expected click-through rates	Yes, through search volume	What an AI answer says, and queries outside the tracked list
AI share of voice	Your share of qualifying brand mentions in a defined set of AI answers (definition and formula)	No, as defined here: every prompt in the set counts equally (some vendors add modelled interest estimates)	How many buyers ask each prompt, or whether the prompt set resembles real demand

The Semrush description is in its guide to measuring SEO share of voice. Plain rankings answer whether a page can be found for a keyword. They do not say whether anyone is looking.

Why one signal cannot stand in for the other

Because the two result sets only partly overlap, a good ranking does not tell you what an AI answer will say, and an absent ranking does not rule out being cited. The measured overlaps differ by study, engine and date, and none is a finding that ranking causes or fails to cause citation:

• Ahrefs' March 2026 AI Overview study found that 37.9% of the URLs cited in Google AI Overviews also appeared within the first 10 blocks of the regular results page for the same query (37.1% counting only standard blue links); the rest sat at positions 11 to 100 or beyond the top 100. Ahrefs' July 2025 study found a higher overlap, but Google's model and Ahrefs' parsing changed, so the two are not like-for-like. Ahrefs measured overlap between two result sets, not whether ranking causes citation.
• Ahrefs' 15,000-prompt overlap study found that, on average, only 12% of the links cited by ChatGPT, Gemini and Copilot appear in Google's top 10 results for the same prompt; Perplexity is the outlier, at nearly 1 in 3. Ahrefs measured whether cited links also rank for the same prompt, not whether ranking causes citation.
• Moz's 2026 analysis of 40,000 queries found that 88% of Google AI Mode citations are not in the organic top 10 for the same query. The figure is the overlap between those two result sets.

One mechanism is documented. Google says AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources. Plausibly, a page can be retrieved through a related search without ranking for the keyword you track. OpenAI says ChatGPT may rewrite a prompt into targeted queries when it searches. Engines differ in how they search, which may be one reason a ranking read and an AI read of the same question can disagree.

One route-specific observation fits alongside those statements. In the answer runs we collect through the API between October 20 and October 31, 2026, our request requires the ChatGPT-model route (gpt-4.1-mini) to use web search, and on every successful run it ran exactly one search whose text was the full prompt. The Claude-model route (claude-haiku-4-5) was free to decide: it searched on most runs, usually with differently worded queries, and only a handful of its queries matched the prompt text. That is one collection route in one window, with prompts we wrote. It is not a description of the ChatGPT or Claude apps, and it does not contradict OpenAI's account of its product. For the ChatGPT-model route, check the full-sentence ranking as well as the short keyword. The logs show the query text, not what the search backend then does with it.

How do you read demand honestly with both?

Take the questions from demand evidence, then read rankings and AI share of voice on those same questions, one surface at a time. This is our method, not a measured result.

1. Source the questions from people actually asking. Use Search Console queries, sales-call and support questions, and the exact wording buyers use. A prompt library is a sample you wrote, not observed demand.
2. Weight them. Use search volume or the count of times a question reaches you. AI engines publish no prompt volume, so the weight has to come from outside the AI answers.
3. Read AI share of voice on the weighted set, repeating each question and counting by the rules in the measurement contract. Report the engine route, the dates and the counts.
4. Read the ranking for the same question text next to it, using the full sentence as well as the short keyword.
5. Diagnose by the pattern, as below.

Ranking	AI answers	Where to look first
Low or absent	Absent	Retrieval: does a page exist that matches the question's wording, and can the engine reach it?
High	Absent	Answer composition and evidence: is the page the kind of source the answer draws on, and does the answer name brands at all?
Low or absent	Present	Other routes: related searches, or third-party pages that mention you. Record which sources the answer cites.
High	Present	Keep monitoring both, and check that the answer describes you correctly.

Machine Relations is the frame for this. It treats being found, being cited and being described accurately as one discipline, so the two signals are two checkpoints on one path rather than rival metrics. The definition lives at machinerelations.ai. Start your visibility audit → to see where your buyers' questions break on that path.

Frequently asked questions

Is AI share of voice a better demand signal than keyword rankings?

No. Neither reads demand on its own. Plain keyword rankings carry no search-volume weight. As defined here, AI share of voice counts every prompt equally; some vendors, such as Ahrefs Brand Radar, add modelled prompt-interest estimates. Use search-demand data to choose and weight the questions, then read both.

Do keyword rankings still matter for AI answers?

They still show whether a search engine can retrieve your page for a question, and some collection routes search the full prompt directly. The measured overlap between rankings and AI citations is partial, so a ranking is an input worth reading, not a prediction.

Can I use AI share of voice to estimate how many buyers ask a question?

No. It measures how often brands appear in answers to prompts you ran, not how often buyers ask them. A prompt set you wrote is a sample of questions, not observed demand.

Method note. The Semrush description of SEO share of voice was read on November 1, 2026 and is a vendor's definition of its own metric. The overlap studies measure how often cited links also rank for the same query, on different engines, dates and sample sizes, and are not comparable with one another. The observation about ChatGPT-model and Claude-model search queries comes from our own API answer-run logs for October 20 to 31, 2026. We have not published the counts, so treat it as a pointer to check in your own tracker's logs, not as a result. This page is not a measurement of any brand's citations.