AI Visibility for SaaS: Prompts, Pricing and Access

SaaS buyers use AI assistants at two moments: building a shortlist, and checking a detail before booking a demo. Here are hypothetical examples, with placeholders where brand names would go:

1. What's the best project management software for a 20-person design agency?
2. Alternatives to [Competitor] that are cheaper for small teams
3. [Your product] vs [Competitor] for a Series A startup
4. Which help desk tools integrate with both Slack and Jira?
5. Does [Your product] include SSO on the base plan?
6. Cheapest CRM with an open API under $30 per user
7. Is [Your product] SOC 2 compliant?

The first four are shortlist prompts. The engine picks a handful of names from a crowded category. The last three are fact prompts. The engine states something specific about one product, and a wrong answer there can end a deal before sales hears about it.

The prompt a buyer types isn't the query the engine runs. OpenAI says ChatGPT search typically rewrites a prompt into one or more targeted queries. Track a fixed prompt set anyway, but treat it as a proxy. A reproducible prompt testing protocol keeps month-to-month comparisons honest.

As of October 2026, five surfaces matter for most software categories. They don't share an index, a model or a citation format, so measure each one on its own.

• ChatGPT searches the web at the model's discretion. When it doesn't search, it answers from training data with a fixed cutoff. OpenAI's model documentation lists February 16, 2026 for the GPT-5.6 family and August 31, 2025 for the Instant path. A plan you launched this spring may not exist in an unsearched answer.
• Perplexity answers from its own index, which PerplexityBot builds.
• Claude decides when to search, and on Claude.ai web search is a user toggle. Tracking Claude's answers means recording that state.
• Google AI Overviews and AI Mode draw on Google's index. Google says a page must be indexed and eligible for a snippet to appear as a supporting link, with no additional technical requirements.
• Microsoft Copilot grounds on Bing. In Otterly's 2026 experiment, Copilot kept citing two sites after Google deindexed them, while ChatGPT barely cited them even though they ranked in Bing.

What it means: a single blended score hides which engine is the problem. The next move is to run the same prompts on each engine and note whether each answer searched.

For shortlist prompts, engines often cite pages other than yours: software review directories, best-tools listicles, head-to-head comparison posts and community threads. When an engine cites other sites instead of yours, being named accurately on those sites is the route into the answer. The evidence shows an association there, not a promised effect.

Find these pages by reading the citations engines return for your shortlist prompts. Then check each one:

• Are you listed in the category the buyer asked about?
• Is your pricing current, including plan names and per-seat or usage limits?
• Are your integrations listed correctly, with retired ones removed?
• Does a comparison table show your old price next to a competitor's current one?

Fix what you control first, which is your own directory profiles. For editorial listicles, send the author the corrected facts and a page that states them. Paying for placement isn't a route to recommendations either, as whether you can pay to appear in ChatGPT explains.

No published study counts which facts AI gets wrong about software companies. These are the facts to check first, because they change often and SaaS sites often hide them from crawlers:

• Plan tiers and names. You renamed a tier last year. Training data may not know.
• Prices and billing basis. Monthly versus annual, per seat versus per workspace, prices shown only after a toggle.
• Limits. Seats, contacts, API calls, storage, projects per plan.
• Feature gating. Whether SSO, audit logs or API access sit on the base plan or only on enterprise.
• Integrations. Native versus via a connector, and retired integrations still described elsewhere.
• Compliance. SOC 2, HIPAA and data residency claims.
• Free plan or trial. Whether one exists, and its terms.

These go wrong for two reasons. A retrieval-off answer is frozen at the model's cutoff. A retrieved answer can only use what the crawler could read. Check your own site for contradictions too: the pricing page says one limit, the docs say another, an old blog post says a third.

The fix is to state each fact plainly in a page that can be fetched. A specific sentence, such as the Team plan costs $20 per user per month and includes 10,000 API calls, gives an engine something it can quote. The 252,000-trial factorial study found explicit price information helped consistently, though it measured pages already placed in the model's context. The SaaS pricing page playbook covers layout, and whether to publish prices at all covers the trade-off.

Access is where the evidence is strongest. Treat each gate as something to check, not assume.

Rendering

Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. Independent retesting in June 2026 found the same for OpenAI, Anthropic, Perplexity and Meta crawlers. Googlebot renders, so Google is the exception. Fetch your pricing page with JavaScript disabled. If plans, prices or the annual-billing figures are missing from the raw HTML, server-render them.

robots.txt, per bot

Rules apply per bot, and OpenAI says each setting is independent. GPTBot governs training only. OAI-SearchBot controls ChatGPT search. A blanket AI block removes search eligibility along with training. A minimal allow for the retrieval bots looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Firewall and CDN

A firewall or CDN can block a crawler that robots.txt allows, often with a 403. OpenAI's help centre says eligibility needs both: allow OAI-SearchBot, and let traffic from its published IP addresses through. From September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and free-tier accounts, while Search stays allowed. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Also look for accidental suppression. A noindex, nosnippet, max-snippet:0 or data-nosnippet inherited from a template keeps content out of AI answers.

Passing every check buys eligibility. It doesn't buy a recommendation.

• Whether an engine searches for a given prompt is its decision.
• Which pages it retrieves, and in what order, sits with the retrieval operator.
• Which retrieved pages it cites is the model's choice. A July 2026 critical survey of 45 GEO studies found no technique with a stable, cross-platform causal effect on discoverability.
• Brand prominence is the strongest association in the evidence, and it doesn't move in a quarter.

So skip the tactics sold as shortcuts. Schema, llms.txt, AI-friendly rewriting and chunking have no credible evidence of getting pages cited, and Google says its AI features need none of them. Mass-publishing comparison pages carries real risk: in Otterly's test, Google removed two 1,000-post AI sites from its index with no manual action reported. For the wider category question, see how to get recommended when people ask ChatGPT for the best tools.

Week 1: baseline. Write 10 to 15 prompts that mix shortlist and fact types. Use your category, your top two competitors and your top three integrations. Run them on each engine, logged out, with memory off. Record whether each answer searched, whether you were named, and every fact stated about you. The free checker on this site gives a quick snapshot for up to three buyer questions.

Week 2: access. Fetch your pricing, plans, integrations and security pages with JavaScript off. Check robots.txt for OAI-SearchBot, PerplexityBot and Claude-SearchBot. Request a page as each crawler to catch firewall blocks. Search templates for snippet directives. Confirm indexing in Google Search Console and Bing Webmaster Tools separately.

Week 3: facts. Give each fact class one canonical, server-rendered page. Resolve contradictions between pricing, docs and old posts. Update your directory profiles.

Week 4: third parties and re-test. Send corrections to the authors of pages cited in week 1. Re-run the same prompts with the same method and compare per engine. Expect retrieval-off answers to stay stale until the next model release.

Why does ChatGPT still show our old pricing? Two likely causes. If the answer didn't search, it came from training data with a fixed cutoff, and no site change reaches it until a new model ships. If it did search, check whether your current prices are in the raw HTML before JavaScript runs, and whether a cited third-party page still lists the old plans. Fix the page and the listings, then re-test with search visible in the answer.

Should a SaaS company block GPTBot? That's a training decision, not a visibility one. GPTBot governs whether content may be used to train OpenAI's models. OAI-SearchBot governs ChatGPT search, and OpenAI says each setting is independent. You can block GPTBot and still allow OAI-SearchBot. What you shouldn't do is apply a blanket AI block that removes the search bots too, because that ends ChatGPT search eligibility.

Do review site profiles get a SaaS product recommended? No study shows that cleanly. The review-profile studies in circulation don't define their metrics and are confounded by brand size. What holds up is narrower: when engines cite review or comparison pages for your category, being listed there with correct facts is the route into those answers. Keep profiles accurate and complete, and treat any lift as possible, not promised.

How often should we re-check AI answers about our product? Monthly is a reasonable rhythm, plus after any pricing change, plan rename or major integration launch. Use the same prompts, the same logged-out setup and the same recording method each time, or the comparison means little. For B2B specifics on citations, see [how a B2B SaaS company can get cited more often](/questions/b2b-saas-cited-chatgpt-perplexity).