Building a Business Case for AI Visibility Work

The honest case for AI visibility work is defensive. You pay to find and remove the things that stop AI engines from reading your site or describing you correctly. That's a smaller claim than most GEO pitches make. It's also the one that survives a finance review.

The research explains why. A July 2026 critical survey of 45 GEO studies concluded that content already retrieved by a model can change how it gets cited, but no reviewed technique shows a stable, cross-platform effect on discoverability. Our digest of that survey walks through the details. The widely quoted "up to 40% more visibility" figure came from a lab engine with the sources already loaded into the model's context. C-SEO Bench went further and found most conversational SEO methods largely ineffective, and often harmful.

So when an agency or vendor hands you a projected citation lift, ask how it was measured. The same questions you'd ask of GEO case studies apply. If your CFO checks that number in six months, there will be nothing to check it against.

The next move: pitch the spend as an audit, a fix list and a re-check. Report what was broken and whether it's now fixed.

Risk 1: AI crawlers cannot reach your content

Access comes before everything else. If a crawler can't fetch a page, no engine can use it. Four common blocks:

• Wrong robots.txt rule for the wrong bot. Rules are per bot. OpenAI documents that each setting is independent: GPTBot governs training use, while OAI-SearchBot controls whether you appear in ChatGPT search. A blanket AI block can remove you from ChatGPT search without anyone deciding to.
• A firewall or CDN that turns crawlers away. OpenAI's help centre says eligibility needs robots.txt permission and a host or CDN that allows its published search bot IP addresses. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. A site can lose access without anyone editing it.
• Content that only exists after JavaScript runs. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If your pricing renders client-side, those crawlers never see it.
• Accidental suppression. Google says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode. A stray noindex, nosnippet, max-snippet:0 or data-nosnippet inherited from a template removes content from AI answers.

Risk 2: AI answers state wrong facts about you

AI platforms get specific facts wrong about specific brands: an old price, a retired product, the wrong location. A buyer who reads a wrong price before visiting your site is a concrete commercial risk. You can't control what a model says. You can check what it says, fix contradictions on your own site, and state the correct fact plainly on a page crawlers can fetch. That gives the engine a sentence it can quote.

1. Pick 5 to 10 buyer questions. Use the questions a prospect asks before they buy, such as "best [category] for [use case]" or "how much does [your product] cost". The free checker on this site runs up to three of them and shows a snapshot of what AI engines say.
2. Record every answer and mark the wrong facts. Compare each answer with your current pricing, products and company details. Note whether the answer showed citations, because an answer without them may come from the model's training data, not a live search.
3. Run the access checks. Read your robots.txt per bot. Request key pages as each AI crawler to see whether your firewall returns a 403; the AI Crawler Checker does this. Compare the raw HTML of your pricing and product pages with the rendered page.
4. Check index status engine by engine. Look at Google Search Console and Bing Webmaster Tools separately. One being healthy tells you nothing about the other.
5. Turn each finding into a line item. Write what is broken, who fixes it, the hours involved and what it puts at risk. "OAI-SearchBot gets a 403 from the CDN" is a line item. "Improve AI presence" is not.
6. Set the success measure and a re-check date. Success is every line item fixed and re-verified by a second check. Put the re-check in the calendar before you ask for approval.

• robots.txt reviewed for OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, bingbot and Meta-WebIndexer, with training bots (GPTBot, ClaudeBot, Google-Extended) treated as a separate decision
• Each AI crawler tested against the live site, with status codes recorded
• CDN or firewall bot settings reviewed, including any defaults that changed in September 2026
• Pricing, product names and key facts confirmed present in raw HTML before JavaScript runs
• Templates scanned for noindex, nosnippet, max-snippet and data-nosnippet
• Google and Bing index status confirmed for your most important pages
• Contradictions on your own site listed (pricing page versus footer versus about page)
• Wrong facts in AI answers listed, with the engine and the date of each answer
• Each item has an owner, an hour estimate and a re-check date

If a line item can't be verified as fixed, it doesn't belong in the case.

Good results

• Every AI crawler you intend to allow gets a 200 response from your live site.
• Your robots.txt and your firewall agree about which bots get in.
• The facts buyers care about appear in the HTML a crawler receives.
• The wrong facts you logged are corrected on your own pages, and you've re-checked the AI answers.
• You can show a dated before-and-after list of what was broken and what is fixed.

Bad results

• A report built around a single blended "AI visibility score" with no breakdown by engine.
• Line items such as "add FAQ schema" or "publish an llms.txt" sold as citation drivers.
• A projected traffic or citation number with no method behind it.
• An "AI ranking position" for Google, which has no data source to verify it.
• Fixes that shipped but were never re-tested.

A bad result isn't a failed audit. It's a case that can't be checked.

Say these limits out loud in the proposal. It protects you when someone asks for ROI later.

• Clicks from AI Overviews. As of October 2026, Google's generative AI report in Search Console shows impressions only: no clicks, no CTR and no position. Nobody can report traffic from AI Overviews.
• Rank inside an AI answer. No engine publishes position data for generative answers, so any rank number has nothing to be checked against.
• Citations as an outcome. Fixing access makes you eligible. It doesn't make an engine pick you. What the model says stays outside your control.
• Answers from training data. When an engine answers without searching, it uses what the model learned before its knowledge cutoff. Your fixes reach those answers only when a future model is trained.
• Markup as a lever. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Google says no special markup or llms.txt is needed for its AI features. Existing markup does no harm, but don't budget for it as a citation driver.

What stays in your control: crawler access, rendering, snippet controls, index eligibility and the accuracy of facts on your own site. That list is your scope.

Access breaks quietly. A CDN default changes, a template ships a noindex, a redesign moves pricing into client-side code. So treat this as recurring work, not a one-off project.

• Monthly: re-run your buyer questions and the crawler access test. Log any new wrong facts.
• Quarterly: run the full checklist, including rendering and index status.
• After any trigger event: a site deploy, a CDN or firewall change, a pricing change or a product rename.

Most of the cost is internal hours: an engineer to fix access and rendering, and a marketer to correct facts. Tooling is the smaller line. Our guide to AI SEO audit tools covers which features matter. If you're deciding whether to hire outside help, read whether a small business needs an AI search service. Once access is fixed, the practical order of GEO work shows what comes next.