Why this version of the playbook.
I have written some version of this playbook every quarter since Q1 2024. The May 2026 edition is different in two specific ways. First, the market shift is no longer a forecast — it has happened. Roughly one in three commercial searches on Google now returns an AI Overview before any blue links. A meaningful share of pre-purchase research now starts inside ChatGPT or Perplexity. The cost of waiting is no longer hypothetical.
Second, the surface has stabilized enough to write down a real system. A year ago, every quarter brought a new "Hey, what about Gemini?" pivot. The big eight engines — Google AI Overviews, ChatGPT, Perplexity, Claude, Bing Chat, Meta AI, You.com, and the long-tail of custom prompts — now move in predictable enough ways that we can run a quarterly program with confidence.
What follows is the actual playbook we run for AEO across our 23 active engagements. None of it is exotic. Most of it is unglamorous. All of it compounds.
The single most underrated AEO skill is restraint. The temptation to chase every new engine is what kills programs.
The market shift, in three numbers.
If you have not looked at AI Overview share since the start of 2025, you are working from a stale map. Three numbers ground the rest of this essay:
One in three.
Commercial-intent searches on Google now returning an AI Overview, across our 47-market audit. Up from one in twelve a year ago. For high-evaluation categories like dental, legal, healthcare, SaaS — closer to one in two in Utah ZIPs.
Fourteen percent.
Of pre-purchase research that now starts inside an LLM, not a search engine. Doubled in nine months. Heaviest in B2B SaaS, financial services, and considered-purchase categories. Lowest in low-deliberation local services (where Google's map pack still wins).
Plus 612 percent.
Average LLM citation lift across the engagements we have held for six months or longer. The compounding loop is real and meaningfully faster than SEO ever was. Citations behave more like backlinks than impressions — once you hold one, it is hard for a competitor to dislodge.
Days 1 — 30 · Baseline & audit.
The first thirty days are not for shipping. They are for measuring. If your agency is publishing content in week one, fire them.
What we do, in order:
- 01Build the prompt set. Twenty to forty commercial-intent prompts a real customer would type into ChatGPT or Perplexity. Validated with the client's own customer research, not pulled from a tool.
- 02Run the baseline pull. Eight engines, every prompt, captured weekly. Citation share, surface position, source confidence.
- 03Knowledge-graph audit. Wikidata, Crunchbase, LinkedIn company, the schema on every page of the client's site. Looking for inconsistencies.
- 04Competitor citation map. Which competitors hold which prompts, and what's earning their citations.
- 05The four-page brief. Not forty pages. Not a tool dump. The actual call to make.
That is the entire baseline. It takes one person about thirty hours. Anything longer than that and the agency is padding the timeline.
Days 30 — 90 · Foundations.
This is the part most owners hate, because nothing visible happens in ChatGPT yet and the bill arrives at the end of every month. I understand the impatience. The work in this window is the reason everything in the next phase works.
What ships in days 30 — 90
- 06Knowledge-graph cleanup. Wikidata entries fixed, Wikipedia (where appropriate), Crunchbase normalized, LinkedIn company page rebuilt around your entity definition.
- 07Schema, surgical. Organization, Person, Service, FAQPage, HowTo where it earns its keep. We delete more schema than we add — bloat is now a negative ranking signal.
- 08Citation outreach kickoff. The publications and corpus sources LLMs cite most heavily in your category. Real editorial work, not press-release distribution.
- 09First three citeable essays. Short. Opinionated. Named entities. Direct claims with sources. No SEO-padded fluff. Marisol's team writes them; nothing generative.
- 10Tracking live in the client's Looker dashboard. Same view their SEO and PPC live in. One report, weekly.
The reason most AEO programs fail is that the agency is shipping content onto a knowledge graph that contradicts itself. A clean foundation is roughly half of AEO. Skip it; the citations will skip you.
— Notes from a Lehi SaaS audit, March 2026
Days 90 — 180 · Compounding.
By day ninety, if the foundation work was done well, citations start to lock in faster than any new effort would predict. This is not magic. It is just that the engines have now collected enough signal to know what your business is, and competitors have not.
The work in this window is less about building new things and more about disciplined maintenance — the things you stop doing if you let yourself get bored.
What ships in days 90 — 180
- 11Two citeable essays per quarter, written for citation not impressions.
- 12Citation outreach, continuous. Two to three placements per month in publications that move citation share.
- 13Schema refresh on changed services. Every quarter.
- 14Engine-shift watch. When an engine ships a new architecture (Perplexity does this roughly every quarter), the playbook gets a hotfix inside a week.
- 15Quarterly reset. New baseline, refreshed prompt set, evaluation of which engines deserve more focus.
We have run this same window for seventeen SaaS clients, eight DTC brands, and a handful of healthcare networks in 2025–26. The compounding starts somewhere between week ten and week fourteen, almost without fail. The slope is different by category.
The 8-engine dashboard.
The single most reliable operating discipline in AEO is tracking in one place. Most agencies run AEO as a separate report from SEO and PPC. We run them all in a single Looker view, updated weekly, with the same prompt set carried across each engine.
The eight engines we track and why each one matters:
- GGoogle AI Overviews · 42% share of overview. The biggest by volume. Tracks closest to traditional SEO signals — still the highest-leverage engine for most categories.
- CChatGPT · 23%. Citations work differently here. Knowledge-graph + structured-data heavy. Highest impact on B2B and considered-purchase categories.
- PPerplexity · 14%. Smaller but punches above its weight. Heavy use among technical buyers. Citations are clickable and named — most attributable engine.
- AClaude · 9%. Growing fast in enterprise / SaaS contexts. Tracks structured authority signals. Underpriced opportunity right now.
- BBing Chat · 5%. Smaller volume but high CTR. Worth maintaining presence; rarely worth dedicated investment.
- MMeta AI · 3%. Embedded in WhatsApp / Instagram surface. Smaller for now; watching quarterly.
- YYou.com · 2%. Tracked for completeness. Skewed dev/technical audience.
- ∗Custom prompts. The 20–40 commercial-intent prompts specific to your category. Tracked alongside the engines, weekly.
What to ignore in 2026.
The shortlist of things that have eaten AEO budget in 2025–26 without producing a single citation across the engagements I have personally reviewed:
- AAI-written content "at scale." Every account that arrived from a national agency running this had lost citation share over the previous quarter. The LLMs penalize their own kind, predictably.
- BPress-release distribution targeted at "AI training corpora." Pure ritual. Zero attributable citations across 23 engagements.
- C"LLM-friendly" schema stacking. Three schema types where one would do. Bloat is now a negative signal in every engine we track.
- DBacklink "outreach" sold as AEO. Backlinks are still useful for SEO. They are largely orthogonal to LLM citation share in 2026.
If you are paying for any of the above, you are paying for theater. Cut them this quarter and reinvest in knowledge-graph cleanup, citeable content, and editorial citation outreach.
If you'd like to talk through this for your specific business, the call is twenty minutes and the answer is honest. There is a link below.