Which weaknesses in my brand's AI search presence can I actually fix?

Every AI visibility problem sits in one of four stages. Only some of them are yours to change.

Access is binary and mechanical, so fixes there are checkable. Retrieval responds to topical relevance and to what other sites say about you, but nobody holds a direct lever. Parametric knowledge and generation belong to the vendor.

A critical survey of 45 GEO studies (arXiv, July 2026) found that content already retrieved can change how it gets cited. It found no technique with a stable, cross-platform effect on being discovered in the first place. So put effort where the mechanism is documented: access first, then facts.

Per-bot robots.txt

robots.txt rules apply per bot. OpenAI states that its crawler settings are independent: GPTBot governs training, and OAI-SearchBot governs ChatGPT search. A blanket AI block can silently remove you from ChatGPT search. This setup stays out of training and stays in search:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Check PerplexityBot, Claude-SearchBot and Meta-WebIndexer the same way. Meta is the one vendor that says allowing its indexer helps it cite you, covered in getting cited in Meta AI.

Firewall and CDN blocks

A firewall can return a 403 to a crawler that robots.txt allows. OpenAI's help center says eligibility needs both: allow OAI-SearchBot, and let traffic from its published IP ranges through. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. You can lose access without editing anything.

JavaScript rendering

Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but don't execute them. If your pricing only appears after JavaScript runs, those crawlers never see it. Googlebot renders, so this mostly affects non-Google engines.

Snippet controls and Google's opt-out

noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. Bing confirmed in October 2025 that data-nosnippet content is excluded from AI summaries. Templates often apply these by accident. Google's Search Console control for generative AI features includes properties by default, but child properties inherit the parent's setting.

Developer tickets for AI access fixes turn each finding into a task an engineer can close. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Contradictions on your own site

If your pricing page, about page and footer state different numbers, an engine can repeat any of them. Pick one version of each fact: price, founding year, location, product names. State it plainly in HTML that loads without JavaScript. A specific fact on a fetchable page gives the engine a sentence it can quote. That's the only on-page signal with positive experimental support, and it still guarantees nothing.

Wrong facts on other people's pages

When engines cite directories, review sites or comparison articles instead of you, those pages shape the answer. The evidence shows a strong association between being named on cited sites and appearing in answers, with no promised effect. Correct the listings you control and ask publishers to fix errors. Fixing ChatGPT errors separates mistakes from training from mistakes from search, and improving AI visibility without website changes covers the off-site work.

Manipulated third-party content

Some pages carry hidden instructions aimed at AI search agents. Research on agent manipulation (arXiv, June 2026) found attack success varied widely by model. Treat it as a reputational risk to monitor. Never copy it.

Frozen training knowledge

With search off, a model answers from what it learned before its cutoff. As of OpenAI's model docs (checked August 2026), the GPT-5.6 family has a February 16, 2026 cutoff and the Instant path an August 31, 2025 cutoff. Nothing you publish today reaches those answers until a new model ships.

Context position

C-SEO Bench (NeurIPS 2025) found that a document's position in the model's context mattered more than rewriting tactics. The engine sets that position. No publisher can invoke it.

Which model answers, and whether it searches

As of October 2026, ChatGPT searches at its own discretion, Claude.ai search is a user toggle, and Google AI Overviews often don't trigger. The same question can get a grounded answer one day and a memory answer the next. That's why how long AI visibility work takes depends on which stage you fixed.

Schema, llms.txt, chunking and AI-friendly rewrites are sold as fixes. The evidence doesn't back them for AI answers.

• Google says structured data isn't required for generative AI search. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema.
• Google Search ignores llms.txt, and no other engine documents reading it.
• The retrieval operator chunks pages inside its own index, as the chunking entry explains.
• C-SEO Bench found most conversational SEO rewrites ineffective and often harmful.

Existing markup does no harm, so don't rip it out. Just don't count it as a fix. The comparison of rewrites and access fixes sets the studies side by side.

1. Read your robots.txt bot by bot. Confirm the search crawlers are allowed, whatever you decide about training bots.
2. Request a page as each AI crawler and compare the status codes with what robots.txt says.
3. Fetch key pages with JavaScript off. Note any price, product or claim that disappears.
4. Search templates for noindex, nosnippet and data-nosnippet, and check Google's generative AI control in Search Console.
5. List your core facts and fix every page that contradicts them.
6. Then check what engines actually say. The free checker on this site shows whether AI engines name you for up to three buyer questions, scored per engine.

Can I correct a wrong fact an AI engine states about my brand? Partly. If the error comes from search, fix the source: your own page, or the third-party page the engine cites. If it comes from training data, nothing changes until a newer model with a later cutoff ships. Make sure the correct fact sits in plain HTML on a page crawlers can reach, so a grounded answer has the right sentence to quote.

Does allowing GPTBot help me appear in ChatGPT? No. OpenAI documents GPTBot as a training crawler only. ChatGPT search uses OAI-SearchBot, and OpenAI says each setting is independent. You can block GPTBot and still appear in ChatGPT search, as long as OAI-SearchBot is allowed and your firewall lets it through.

Will fixing access guarantee my brand gets cited? No. Access fixes remove blocks. They make your pages eligible to be fetched and considered. Whether an engine retrieves and cites a page for a given question depends on relevance, competition and the model, none of which you control. Access is still the right first step, because a blocked page can't be cited at all.

How do I tell which stage a problem sits in? Start with access: if a crawler gets a 403 or sees an empty page, the problem is access. If access is clean but answers are wrong only when the engine doesn't search, it's parametric. If grounded answers cite other sites with wrong facts, work on those sources. Measure each engine separately, because they behave differently.