The complete AEO strategy playbook

Most "AEO strategy" content treats every AI engine as one undifferentiated blob and every industry as interchangeable — swap in "accounting firms" or "legal firms" and republish the same five bullet points. It doesn't work, because the engines don't work the same way underneath, and a compliance-bound law firm has almost nothing in common with an ecommerce brand shipping SKUs. This guide breaks both apart properly: what actually differs in how ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot, and AI Overviews decide what to cite, and what actually differs in how accounting firms, legal firms, and ecommerce brands need to execute against that.

How each AI engine actually sources and verifies answers

"Optimize for AI search" is not one task. The seven engines below pull from different indexes, weight different trust signals, and reward different content shapes. Building for one and assuming the rest follow is how a page ends up cited by Perplexity and invisible everywhere else.

ChatGPT

When ChatGPT's search mode is active, it issues live queries against Bing's index and reads a small set of results, favoring pages with a direct-answer opening sentence, clean H2/H3 structure, and explicit FAQ blocks it can lift verbatim. Outside of search mode, its base answers are shaped by pretraining exposure — pages that were widely quoted, linked, or reproduced across the web by the model's training cutoff have an outsized chance of surfacing in an unsearched response. Getting cited means winning both games: structure pages so live retrieval can extract them cleanly, and get quoted by sites (review platforms, forums, trade press) that ChatGPT's underlying index already trusts.

Claude

Claude's web search tool issues targeted queries and reads full pages rather than short snippets, which rewards well-organized long-form writing with the topic sentence stated up front and evidence laid out in order. Anthropic has trained Claude to weigh source credibility and hedge on uncertain claims, so pages that show their own sourcing — named studies, dated data, explicit caveats — get treated as more citable than confident, unsourced assertions. Getting cited means writing like a careful analyst: define the term, state the claim, cite the evidence, note the limits.

Gemini

Gemini is tightly wired into Google's Search index and Knowledge Graph, so it inherits classic Google signals almost wholesale: schema.org markup, entity clarity, and whether a page is already indexed and understood as authoritative on the topic. It has less of an independent retrieval layer than Perplexity or ChatGPT-with-browsing — it largely trusts what Google Search already trusts. Getting cited means doing structured, technically clean SEO first: Product, Organization, and FAQPage schema, a clear entity association with your brand, and topical depth across a cluster of pages rather than one isolated post.

Perplexity

Perplexity runs real-time web search on every query and numbers its sources inline, which makes it the most transparently citation-driven of the seven. It rewards pages that answer the query directly within the first couple of sentences, carry a visible publish or update date, and lead with original data or numbers rather than throat-clearing. It also appears to favor domain diversity — a niche, authoritative source can outrank a generic content-farm repost on the same topic. Getting cited means publishing (and dating) original stats, keeping pages current, and making sure PerplexityBot is allowed in robots.txt — a blocked crawler means zero visibility here regardless of content quality.

Grok

Grok is the outlier: it pulls heavily from real-time posts on X in addition to general web search, and it visibly favors content that is being actively discussed or shared on the platform. A page with no social conversation behind it is competing on web search alone; a page that's being quoted, debated, or linked in active X threads has a second, independent path into Grok's answers. Getting cited means pairing the written page with an active X presence — posting the core claim, linking back, and letting real engagement accumulate — rather than treating the article as a standalone asset.

Microsoft Copilot

Copilot runs on Bing's index, so its dynamics resemble ChatGPT-with-browsing but weighted toward signals Bing specifically tracks: Bing Webmaster Tools verification, crawlability, and Bing Places / LinkedIn data for anything company- or local-related. Because Copilot sits inside Microsoft's own ecosystem, a complete, active LinkedIn Company Page is a real citation input, not just a social nicety. Getting cited means verifying in Bing Webmaster Tools, keeping technical SEO clean (Bing is stricter about crawl errors and stale sitemaps than Google tends to be), and keeping your LinkedIn entity data current.

Google AI Overviews

AI Overviews is fundamentally a summarization layer generated at query time from pages that are already ranking in Google's top organic results — it rarely surfaces a source that isn't on page one or two organically. That makes it the engine most dependent on classic E-E-A-T and technical SEO fundamentals: author and organization schema, comprehensive topic coverage, internal linking, and passages written as self-contained, single-claim paragraphs that extract cleanly. Getting cited means ranking well first, then structuring the page so the winning passage is unambiguous — a clear definition sentence, a direct number, a scoped claim — rather than burying the answer in narrative.

Engine-by-engine comparison table

The table below condenses the mechanics above into a working reference: what each engine sources from, the format it rewards, and the single highest-leverage move for getting cited.

AEO for accounting firms

Accounting content sits inside a hard constraint the engines are specifically tuned to notice: it cannot read as personalized financial or tax advice without qualification, and models trained on regulated-content patterns treat unsourced, absolute claims in this category with more suspicion, not less. That constraint is workable, and in practice it becomes an advantage — a page authored by a named CPA, citing IRS.gov or a state board directly, with an explicit "this is general information, consult a licensed professional for your situation" scope statement, matches the exact trust signature these models are trained to reward in financial topics. A generic "ultimate guide to small business taxes" with no byline and no citations reads as lower-trust by comparison, regardless of how well it's written.

The practical build: publish threshold and deadline content with exact numbers (filing dates, mileage rates, contribution limits) sourced directly to IRS or state authority pages, add Person schema with CPA credentials to author bylines, use ProfessionalService / AccountingService schema on service pages, and structure FAQ sections around the literal questions clients ask ("do I need to file quarterly if I'm a sole proprietor in [state]") rather than marketing-flavored headlines. Engines like Gemini and AI Overviews, which inherit Google's E-E-A-T weighting, respond especially well to visible author credentials and a clean citation trail back to primary sources.

AEO for legal firms

Legal content carries a sharper version of the same constraint, plus one accounting doesn't: unauthorized practice of law risk means content genuinely cannot read as jurisdiction-specific legal advice, and jurisdiction is not optional context — the correct answer to almost every legal question changes by state, sometimes by county. Content that ignores this (a single page claiming to cover "statute of limitations for personal injury" nationally) is both a compliance risk and a worse AEO asset, because it can't state the specific, verifiable number an engine wants to extract. A page scoped to one state, citing the actual statute section, will out-cite a vague national overview every time a query includes that state.

The practical build: structure content by jurisdiction rather than by topic alone (separate, clearly labeled pages or sections per state), link directly to statute text and court rules instead of paraphrasing them, add attorney bios with bar admission and jobTitle/hasCredential schema, and avoid outcome-guarantee language ("we'll win your case") that both violates advertising rules in most states and reads as an unsupported claim to models trained to discount them. Claude in particular — given its training toward hedged, evidence-based answers — responds well to legal content that states what the law says and separately, clearly, states what a reader should do next (consult counsel in their jurisdiction).

AEO for ecommerce brands

Ecommerce AEO is a different problem entirely from the two above: the binding constraint isn't compliance, it's freshness and structure at the product level. "Best noise-canceling headphones under $200" and "is this in stock" are comparison-shopping and availability queries, and engines answering them — Gemini and AI Overviews especially, increasingly Perplexity and ChatGPT's shopping features — pull directly from structured product feeds (Merchant Center, Product schema) rather than from prose. A beautifully written product page with a stale price in its schema will get silently dropped from AI shopping answers; the engine has no way to know the number is wrong until it's already burned trust once.

The practical build: keep Product, Offer, and AggregateRating schema accurate and synced to real-time inventory and pricing, never let a feed go stale longer than a checkout cycle, build genuinely comparative content that names competitor products and specs honestly rather than only your own catalog (engines cross-reference and penalize one-sided "comparisons"), and use real review counts and ratings rather than inflated or fabricated aggregates, which both violate schema guidelines and erode the trust signal engines are specifically checking for on commercial queries. Grok and Perplexity's freshness weighting make ecommerce one of the categories where AEO discipline compounds fastest — and decays fastest when neglected.