Are AI search share of voice optimization packages worth buying?

Most share of voice packages sell a metric, then sell activity to move it. The metric is weaker than it looks, so judge the package by its deliverables instead.

AI share of voice is your mentions or citations divided by all mentions or citations across a set of prompts. Two things decide that number before any optimization starts.

• The prompt set. The vendor picks the prompts. Change the list and the score changes. OpenAI's help page on ChatGPT search says it rewrites your query into one or more targeted queries before sending them to search providers. So a tracked prompt is a proxy for an input nobody outside the engine sees.
• The denominator. Whoever already wins the query fills most of it. The 2025 study of AI search citing patterns (arXiv 2507.05301) found the top 20 news sources accounted for 67.3% of all citations for OpenAI models. Big incumbents set the base you are measured against.

Say you run a hypothetical B2B software brand. In month one you get 10 of 100 tracked citations, a 10% share. In month two the category gets busier: you get 15 of 200. Your citations rose by half and your share fell to 7.5%. A package sold on share would call that a failure.

That is why citation rate and citation share need reading separately. Track your own count first.

Share is a competition metric, and the research says competition erodes gains. C-SEO Bench (NeurIPS 2025, arXiv 2506.11097) found that most conversational SEO methods were largely ineffective and often hurt ranking. It also found that as more sites adopted those methods, overall gains shrank, which the authors describe as congested and zero-sum.

The wider evidence is narrow. The July 2026 critical survey of 45 GEO studies (arXiv 2607.14035) concluded that no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability. Nearly every published GEO effect was measured on pages already inside the model's context.

Prompt samples also undercount. In Otterly's 2026 experiment, Microsoft's AI Performance report credited one test site with 42,817 citations, while Otterly's prompt-based tracker logged 3,934. The time windows may differ, so it is not an exact ratio. It does show that any prompt sample is a sample.

What this means: a vendor cannot honestly promise your share. They can fix what blocks you.

These map to the four stages of an AI answer. Access is where publishers have real control. Retrieval and generation are mostly up to the engine, and the model's frozen training knowledge is out of anyone's reach this quarter.

1. Per-bot crawler access. robots.txt rules are per bot. OpenAI's crawler page says GPTBot governs training and OAI-SearchBot governs ChatGPT search, and each setting is independent. A good package checks PerplexityBot, Claude-SearchBot and Meta-WebIndexer too.
2. Firewall and CDN reachability. A firewall can return a 403 to a crawler robots.txt allows. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
3. A rendering diff. Content that appears only after JavaScript runs may be invisible to most AI crawlers. Ask for a raw-HTML versus rendered comparison of your pricing and product facts.
4. Index and snippet checks. Search Console and Bing Webmaster Tools, checked separately, plus a scan for stray noindex, nosnippet, max-snippet and data-nosnippet.
5. A factual accuracy audit. Which engines state your price, features or ownership wrong, and where that wrong fact likely comes from.
6. Per-engine reporting with a disclosed prompt list. Each platform scored on its own, with prompts, dates and whether search was on.
7. Third-party presence work. When engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer. Treat it as an association, not a promised effect.

For what a fair scope and price structure look like, see AI SEO packages.

Walk away from a proposal that leans on any of these:

• A guaranteed share of voice, citation count or score. Nobody controls what the models say.
• An "AI rank" or position in Google AI. Google's AI reporting exposes impressions only, so no position data exists to check against.
• "Up to 40% more visibility." That figure traces to the original GEO paper, measured in a lab engine on content already in context.
• Schema, FAQ schema or llms.txt as citation tactics. Ahrefs' matched test found AI Overviews citations fell 4.6% after adding schema. Google says Search ignores llms.txt.
• An E-E-A-T score or domain authority building. Google says E-E-A-T isn't a specific ranking factor, and no engine computes Domain Authority.
• Volume publishing. In Otterly's 2026 test, Google deindexed both 1,000-post AI sites with no manual action warning.
• Hidden text or instructions aimed at AI crawlers. Injected content is manipulation. It is a reputational and policy risk, not a service.
• One blended cross-platform score. In the Otterly test, one engine cited the sites thousands of times while four barely did. A blend hides that.

Run your own baseline first, so you can judge any vendor's numbers.

1. Write 10 to 20 buyer questions in your customers' words, and keep the list fixed. The step-by-step share of voice playbook covers the method.
2. Record your own mention count per engine, not just share.
3. Check crawler access and rendering yourself. Most of it is free.
4. Ask each vendor for their prompt list, their retrieval settings and what they will fix in week one.

The free checker on this site gives a snapshot of whether engines name you for up to three buyer questions. If the gaps are technical and your team can ship them, you may not need a package. Whether to hire help or do it in-house depends on who can change your robots.txt, CDN and templates.