Does what works on one AI engine work on the others?

A tactic that lifts you on one AI engine usually tells you nothing about the others. The reason is structural. Each answer passes through four stages, and engines differ at every one:

• Parametric: what the model already knows from training. Each vendor has its own cutoff, and nobody can change it this quarter.
• Access: whether that engine's crawler may fetch and store your page. Every engine has a different crawler.
• Retrieval: whether your page enters the model's context for a given query. Each engine queries its own index, rewrites prompts its own way, and decides per query whether to search at all.
• Generation: which retrieved pages get used and cited. Different models, different citation formats.

So there is no single "AI search" to optimize for. Google even says its own AI Overviews and AI Mode "may use different models and techniques" and can return different links. The research backs this up. The critical survey of 45 GEO studies (arXiv 2607.14035, July 2026) concluded that "no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability." Our digest of the 45-study survey covers how it reached that conclusion.

The quickest way to see why results diverge is to map each engine to its crawler and its control, as of October 2026.

Two details trip people up. First, OpenAI says its bot settings are independent: blocking GPTBot stops training use but does not remove you from ChatGPT search, which runs on OAI-SearchBot. Second, rendering differs. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, while Googlebot renders pages. A JavaScript-only pricing table can be visible to Google and missing for ChatGPT. Our per-engine map of SEO for AI search goes through each one.

OtterlyAI ran the clearest public example in 2026. It published about 1,000 AI-generated posts on each of two new domains and tracked citations across seven engines for about four months.

• Google deindexed both sites. Google's AI surfaces, which need pages indexed, then almost never cited them.
• Bing kept both sites indexed. Of the 3,934 citations Otterly tracked for one site, 2,718 (69%) came from Microsoft Copilot.
• ChatGPT cited the sites effectively zero times, even while they ranked in Bing and ChatGPT's crawlers kept visiting.

The lesson isn't to mass-publish. Google removed both sites without warning. The lesson is that one engine can cite you thousands of times while four others barely notice you. It also undercuts a common shortcut: "ChatGPT uses Bing" did not hold in this test. It is one pair of sites on a low-stakes topic, run by a vendor, so treat it as a strong example rather than a law. The method and limits of the Otterly experiment are worth reading before you quote it.

The levers with real evidence behind them work on every engine, because they remove blocks rather than add tricks. They are access levers, and you control all of them:

• Per-bot robots.txt permission. Allow the search crawlers you want (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot, bingbot). A blanket AI block often removes ChatGPT search eligibility by accident.
• Network reachability. A firewall or CDN can return 403 to a crawler that robots.txt allows. OpenAI says eligibility needs both.
• Server-rendered content. Put prices, product names and key facts in the HTML before JavaScript runs.
• No accidental suppression. noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers on Google and Bing.
• Plain statements of fact. A specific price or one-line definition on a fetchable page gives any engine a sentence it can quote.

What does not carry across is anything sold as an additive GEO technique. Published effect sizes were mostly measured on pages already inside a model's context, and the best-designed studies found those techniques ineffective or harmful. None of this guarantees a citation. Passing these gates makes you eligible. Each engine still decides what to retrieve and cite.

Work engine by engine, in this order:

1. Check access for each crawler. Compare what your robots.txt says with what your firewall actually returns to each bot. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
2. Check index status separately. Look at Google Search Console and Bing Webmaster Tools on their own. A healthy Google index says nothing about Bing, and Bing is what Copilot draws on. The Copilot visibility checklist covers the Bing side.
3. Compare raw HTML with the rendered page. Anything that only appears after JavaScript runs may be invisible to non-Google crawlers.
4. Measure each engine on its own. A single blended number would have hidden Otterly's whole result. See one blended score versus per-platform scores for why. The free checker on this site gives a quick snapshot for three buyer questions, scored per engine.
5. Fix wrong facts at the source. If engines repeat an error, look for contradictions on your own pages first.