Whether ChatGPT, Perplexity, and Google AI Mode need separate strategies is settled one step before any of them: a page whose content only appears after JavaScript runs is an empty page to the AI crawlers Vercel and MERJ tested, which fetch JavaScript without executing it (Googlebot is the exception). The engines do then pull from different pools: Seer found 87% of SearchGPT citations matched Bing's top results in a 500-citation sample, while Ahrefs found only 38% of Google AI Overview citations rank in Google's organic top 10. No public third-party study measures AI Mode's citation set, so the AI Overviews figures here are the nearest available proxy for Google's AI surfaces, not a reading of AI Mode.
Retrievability is the shared gate. The engines do cite different sources, and the work that follows is still one gate (fetchable pages, a resolvable entity, extractable proof) plus per-engine plays that differ mainly in which third-party surfaces each engine reads. Three parallel strategies is two more than the evidence supports. The stakes are identical on every engine: G2 found AI chatbots and review sites are now the two biggest influences on B2B shortlists, and one in three buyers said an AI chatbot led them to choose a vendor they hadn't considered.
Do ChatGPT, Perplexity, and Google AI Mode really cite different sources?
Yes, and the differences come from how each engine retrieves. ChatGPT answers commercial questions by blending what its model already believes with a live search: OpenAI's help documentation says ChatGPT search issues one or more targeted queries when a question needs current information, and the only public sample of where those queries land, Seer's 500 SearchGPT citations, pointed overwhelmingly at Bing's top results. The memory half is real too: OpenAI lists three primary classes of training source, public web content among them, so years of third-party coverage travel with it.
Google AI Mode runs on Google's own index with a different retrieval move. Google states that AI Overviews and AI Mode use a query fan-out technique, issuing multiple related searches and drawing sources from across them, and that pages must be indexed and snippet-eligible to appear. Ahrefs measured the shift away from the organic top 10 on AI Overviews rather than AI Mode, because no equivalent public study of AI Mode exists.
Perplexity's product describes itself as retrieval-first, searching the live web per query and attaching numbered citations; no third-party study cited here measures that. Its source mix gets far less independent auditing than ChatGPT's or Google's, so most percentage charts for it come from tool vendors' own panels. Treat those as hypotheses.
The blended picture is less engine-specific than the charts suggest. Profound's citation research found more than 4 in 10 AI citations point to sources brands do not control, so across the answers it sampled, a large share of citations already sat on pages no company owns. Which surfaces each engine weights is the part that varies.
Why does the same question get a different answer on each engine?
Two causes survive scrutiny: each engine retrieves from a different pool, and each weights its training memory differently. Most of the rest is run-to-run noise, so establish that a gap survives repetition before blaming an engine.
Different retrieval pools produce different candidate pages, and different weightings of training memory produce different prior associations, which a 2026 analysis of brand dynamics in LLM recommendation systems examines.
SparkToro and Gumshoe's research found AI engines are highly inconsistent when recommending brands, and Search Engine Land's write-up of the same study reports that those lists rarely repeat exactly, even on the same engine.
Ronald Sielinski's work on quantifying uncertainty in AI visibility puts confidence intervals on the readings and reports platform medians rather than one blended score, and a 2026 variance-components study of AI visibility measurement decomposes where the variation comes from. Only a gap that holds for weeks on one engine and not another is worth an engine-specific budget line.
What do all three engines demand before any engine-specific tactic pays off?
Four things, in order: the engine can fetch your page, the page is indexed where that engine looks, the engine can tell who you are, and the page carries extractable proof. Fail the first and nothing downstream matters. The trap is that Google hides this failure: Googlebot renders client-side pages through its Web Rendering Service, so a JavaScript-rendered site can rank in Google and appear in AI Mode while reading as an empty shell to the AI crawlers Vercel and MERJ tested.
Access control is the other half of fetchability. OpenAI documents separate user agents for search retrieval and for model training, and a robots.txt rule that blocks the search agent removes your pages from ChatGPT search answers. Check the rules someone wrote in 2024 before spending a dollar on content.
Indexing differs in mechanism while the requirement is shared. For Google, the AI features documentation says there are no additional requirements and no new markup beyond indexing and snippet eligibility. For ChatGPT, Bing's index was the gate in the only public sample of this, so treat it as the best available proxy rather than a current guarantee. A raw-HTML page with a clean canonical is the safest common denominator.
Entity clarity and proof are the last two shared gates, and where content budget belongs. On entities, one name and one description everywhere help engines attribute claims to the right company, though consistent naming will not lift retrieval on its own: in a 2026 study of 443 entity-oriented retrieval configurations on a news benchmark, entity signals added at most 0.051 MAP once they could not peek at the answers. On proof, the GEO study (Aggarwal et al., KDD 2024) found adding statistics, quotations, and citations lifted citation visibility by up to 40% in its benchmark while keyword stuffing did nothing. The later C-SEO Bench work, a different benchmark scored on citation rank rather than share of the answer, found those rewrites mostly produced no significant gains.
Which engine-specific plays are worth running, and which are folklore?
The plays worth running follow from documented retrieval mechanics, and there are fewer than the playbooks suggest. The corrective comes first: C-SEO Bench (NeurIPS 2025) tested ten conversational-SEO rewrite methods across two tasks and six domains and found most of them largely ineffective and often harmful to citation rank, with gains shrinking as more sites adopted them. Most rewrite-your-page-for-engine-X advice lives inside the 51 conditions that did nothing.
For ChatGPT, the defensible plays are index and sources. Be findable in Bing for the buyer's phrasing, since that is where its live search looks, and present in the third-party surfaces it cites. AirOps' March 2026 analysis of 548,534 pages ChatGPT retrieved found only 15% were cited, and pages whose titles overlapped the query by at least half were cited 20.1% of the time against 9.3% for low overlap, though correlation across a page corpus is not a test of what causes a citation.
For Google AI Mode, the play is mostly the SEO you already run plus topic coverage that survives fan-out. Google's own guidance on succeeding in AI search is to build valuable content and make sure Google can access it. What changed on the sibling surface is the payoff: Ahrefs measured a 58% lower average clickthrough rate for the top-ranking page when an AI Overview is present, so ranking first buys fewer clicks than it used to.
For Perplexity, the play is freshness of the fetch, held loosely because the third-party evidence is thin. If its live retrieval works as the product describes, your page is re-read at question time, so an updated page that answers in the first sentence of the relevant section has a mechanical advantage. Do not buy a promised Perplexity citation share; nobody outside Perplexity has the data to sell one.
How should you split effort between the shared gate and per-engine work?
Treat the shared gate as a fixed cost paid once and per-engine work as a variable cost justified by an observed gap. The gate is cheap: a raw-HTML fetch of your key pages, a robots audit, an index check in Google and Bing, and a pass over entity consistency take days. Per-engine work takes longer, because third-party presence has to be earned and engines change their retrieval between quarters.
Sequence per-engine work by where your buyers ask. 51% of software buyers now start research with an AI chatbot more often than with Google, and in Profound's 7.5 million-conversation en-US sample, the share of ChatGPT conversations classified commercial rose from 13.9% to 19.2% in a year; Profound's "more than doubled" volume figure is an extrapolation from estimated weekly users. For most B2B categories that puts ChatGPT first on volume, Google AI Mode second because it inherits your existing SEO, and Perplexity weighted by what your own answer runs show.
The reason to fund any of it is the shortlist. 6sense found buyers evaluate roughly five vendors and nearly four out of five are named to the shortlist on day one of the buying journey. If the list forms inside an answer, which engine wrote it matters less than whether it could find you at all.
How does Trovance separate a shared retrievability problem from an engine-specific gap?
Trovance runs the same buyer questions across engines on a schedule and preserves every answer run with its citations. Because the questions repeat on each engine, the pattern across runs shows whether your absence is shared or engine-specific. You define the tracked questions, and each answer snapshot records who was mentioned, who was cited, who was recommended, and which sources carried the answer. Share of answer is reported per engine rather than collapsed into one score, and the set of engines covered expands as surfaces are added.
A brand missing on every engine while its pages resist a raw fetch has a retrievability problem, and the recommended action is mechanical and shared. A brand present in Google's AI Overviews and absent on ChatGPT, with the ChatGPT citations pointing to third-party pages that omit it, has an engine-specific source gap, and the action names the surface to earn. Because every snapshot keeps its citations, you can compare which sources each engine leaned on.
The Brain keeps the claims you are entitled to make with the proof behind each one. When the diagnosis says a page needs evidence, the draft uses your author voice guidance and research sources, and a person reviews it before it ships. After it goes live, the next analysis cycle reruns the same questions on the same engines and shows whether the answer moved, on which engine, and against which sources.
What Trovance will not do is promise a citation on any engine, or a single visibility number that pretends the engines agree. The engines are probabilistic, they change their retrieval between quarters, and the variance documented above applies to every measurement. What it does instead is keep the evidence current, so the decision about whether an engine-specific play is worth its cost is made against this month's answers, not last year's chart.
What should you do this week?
Run the shared gate first. Fetch your five most important buyer-question pages with JavaScript disabled and read what an engine reads. Open robots.txt and confirm you are not blocking the search agents of the engines your buyers use. Check that those pages are indexed in Google and Bing under the phrasing a buyer would type.
Then measure before you diverge. Run your core buyer questions in two or three phrasings each on ChatGPT, Perplexity, and Google AI Mode, once a week for a few weeks, recording mentions, citations, and recommendations separately, using the IAB's distinction between presence, prominence, portrayal, and persuasion so the readings are comparable. Only a gap that holds across runs on one engine and not the others earns a per-engine play, drawn from the citations you recorded.
Be honest about what the evidence supports. The retrieval gate and the third-party dependence are documented; the tactic list is mostly unproven, and the largest retest found most rewrite methods largely ineffective and often harmful to citation rank. If you want the per-engine reading without hundreds of manual runs, start a free Trovance trial and see how your answers differ across engines over repeated runs.
Engine by engine
Where ChatGPT gets its information about your business - retrieval plus memory, explained.
What Google AI Mode cites and why - fan-out and what the index still decides.
How to get cited on Perplexity - what live retrieval rewards.
How to get featured in Google AI Overviews - the sibling surface's source mix.
The shared gate
AI crawlers don't run your JavaScript - the failure that loses everywhere.
The four gates of GEO - fetch, index, entity, proof.
How to measure AI search visibility without one score - per-engine readings that survive variance.
Google rankings vs AI citations - why ranking first does not guarantee a citation.
