A GEO program runs in six steps: a baseline audit, entity cleanup, content restructuring, earned media targeting, monthly measurement and a 90-day shortlist of fixes. Start with access and accuracy, because an engine cannot cite a page it cannot reach or a brand it cannot identify, then build the independent evidence engines read. This playbook covers the work three ways: the practitioner steps, the executive view of budget and board reporting, and the seven inputs a brand PR team controls. It assumes you know the basics from What Is GEO?.
By the EPR Editorial Team. Updated October 21, 2026. Part of The GEO Canon. The 14-layer audit behind step 1 is The GEO Operating Stack.
What does the GEO playbook look like at a glance?
| Step | Goal | Main output |
|---|
| 1. Baseline audit | Know where the brand is named, missing or described wrongly | A dated Citation Share baseline and a source map |
| 2. Entity cleanup | Make the brand identifiable and consistent | One approved fact sheet used everywhere |
| 3. Content restructuring | Make pages easy to quote | Answer-first pages with sourced numbers |
| 4. Earned media targeting | Get described by sources engines read | A target list of outlets and a research asset |
| 5. Monthly measurement | See whether the work is holding | A repeated prompt run reported as a range |
| 6. 90-day shortlist | Focus on the fixes that matter first | A ranked list of actions with owners |
GEO marketing is the practice of earning a brand's place inside the answers ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews generate when buyers ask category questions. It works upstream of the click, at the moment an engine names a few brands and the buyer's shortlist forms. A survey of 500 US consumers who use AI tools found that 37% begin searches with AI instead of a traditional search engine, so the shortlist increasingly forms before a website is visited.
| Channel | What it competes for | Main lever | Main metric |
|---|
| SEO | A ranked position for a page | Content, links, technical health | Rankings and organic sessions |
| Paid media | An impression or a click | Budget and targeting | Reach, clicks, cost per acquisition |
| GEO marketing | A named place in the answer | Entity clarity, independent coverage, extractable content | Citation Share across a fixed prompt set |
The disciplines share a technical base. Google says the best practices for SEO remain relevant for its AI features and that no additional requirements apply to appear in AI Overviews or AI Mode, but that guidance covers Google's surfaces, not ChatGPT, Claude or Perplexity. The full comparison is in GEO vs SEO.
What are the five components of a GEO marketing program?
| Component | What it means | First action |
|---|
| Entity authority | A verifiable, consistent identity across Wikipedia, Wikidata, LinkedIn, schema markup and owned properties | Align the name, category, founding date and leaders everywhere they appear |
| Earned media | Independent coverage that engines retrieve and cite, treated as retrieval infrastructure instead of an impressions metric | List the outlets engines cite for your category and pitch those first |
| Extractable content | Pages built so an engine can lift a clean passage: answer-first paragraphs, FAQ blocks, comparison tables, numbered steps | Rewrite priority pages so each heading is answered in the first sentence |
| Cross-engine coverage | Testing and optimizing across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, which retrieve and cite differently | Run the same prompts on all five engines |
| Measurement | Citation Share scored on a locked prompt set and repeated on a schedule | Baseline now and re-run monthly |
Step 1: How do you run the baseline audit?
Run a fixed prompt set across the major engines and record what each says. Use 35 to 100 buyer questions across definitional, comparative, procedural and recommendation types, and run each prompt at least three times per engine, because answers vary between runs. Record whether the brand is named, which competitors are named, which sources are cited and how the brand is described.
Check access at the same time. Confirm that robots.txt and your firewall allow the crawlers each engine uses: OAI-SearchBot, GPTBot and ChatGPT-User for OpenAI, ClaudeBot, Claude-SearchBot and Claude-User for Anthropic, and PerplexityBot for Perplexity. Google-Extended controls whether content is used for Gemini training and grounding, and does not affect Google Search or AI Overviews. The per-engine details are in the guides for ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews. The full method is in The AI Visibility Audit and the six-step citation audit.
Step 2: How do you clean up entity data?
Build one fact sheet and make every public surface match it. List the legal name, category, founding date, headquarters, leaders, products, key figures with sources, and an approved one or two sentence description. Then compare it with the website, About page, executive bios, LinkedIn, Crunchbase, Google Business Profile, Wikipedia and Wikidata where an entry exists, and the press boilerplate.
Most damage in AI answers comes from inconsistency, such as an old address on one profile or two founding dates, not from missing keywords. Where a Wikipedia entry is warranted, follow its conflict-of-interest and notability rules, as covered in The Wikipedia Hub.
Step 3: How do you restructure content so engines can quote it?
Rewrite priority pages so each section opens with a self-contained answer. Put the answer in the first 40 to 80 words, phrase headings as the questions buyers ask, and attach a named source to every number. Use a comparison table where the question is a comparison and numbered steps where it is a process. Add a visible update date and a named author, and keep the content present in the initial HTML.
The original GEO paper reported that adding citations, quotations and statistics raised a source's visibility in generative engine responses by up to 40% in its benchmark. Google says no special markup or AI-specific files are needed to appear in its AI features, so spend the effort on clarity, not tricks.
Aim at the outlets the engines already cite in your category, not at the biggest names by default. Use the source map from step 1 and the Who Controls AI Answers Index to list the publications, reference sites and communities that appear in answers for your prompts, and compare your presence there with competitors'.
Muck Rack's May 2026 analysis found that earned media, broadly defined, accounted for 84% of links cited by ChatGPT, Claude and Gemini, while paid and advertorial content was 0.3% of the citations it studied. Ahrefs found that only about 12% of URLs cited by AI assistants rank in Google's top 10 for the original prompt. Together they point to independent coverage and original research, not ranking alone, and they are why paid media does not directly earn a place in AI answers. EPR's own analysis is in 94% of AI Citations Come From Earned Media. Build one research asset journalists can cite and pitch it to the outlets on your list. Avoid paid or fake mentions, which Google's guidance says may be treated as spam.
Step 5: How do you measure monthly?
Repeat the baseline prompt run every month with the same prompts, engines and method, and report a range. Track Citation Share, mention frequency, share of AI voice against competitors, sentiment and the sources cited. EPR's weighted formula is in Measuring GEO. Cited sources rotate quickly: EPR's citation half-life research found that only 10.6% of cited URLs persisted across 28 days, and that 40 to 60% of cited sources rotated monthly. A single placement or page does not hold a position, and a quarterly rhythm misses the movement.
Step 6: What goes on the 90-day shortlist?
Rank the fixes by how much they unlock and how fast they ship, and give each an owner.
| Window | Focus | Examples |
|---|
| Days 1 to 30 | Access and accuracy | Fix crawler blocks, publish the fact sheet, correct wrong profiles, run the baseline |
| Days 31 to 60 | Pages and sources | Rewrite the ten pages that answer the highest-intent questions, finalize the outlet target list, start the research asset |
| Days 61 to 90 | Evidence and re-measure | Publish and pitch the research, place expert commentary, repeat the prompt run and report change against baseline and competitors |
Which five skills does a GEO practitioner need?
- Prompt and measurement design. Building fixed prompt sets and reading variable answers honestly.
- Technical access. Reading robots.txt, firewall rules and rendering, so engines can reach the pages.
- Editorial writing. Producing answer-first, sourced pages that stand on their own.
- Media relations. Earning coverage and expert mentions in the outlets engines read.
- Entity and reference management. Keeping facts consistent across profiles, Wikipedia and structured data.
Why does GEO belong with communications and not only with SEO?
Most of the inputs that shape AI answers are communications outputs: earned media, owned thought leadership, structured brand entities, Wikipedia and authoritative third-party content. The skills the work needs, such as narrative architecture, source cultivation, message discipline and executive positioning, sit in PR more than in paid digital. Teams that do this well treat GEO as a strategic communications layer, with search teams handling technical access, and they give AI visibility a named owner. The function does not need a big budget. It needs a clear owner.
The executive view: how does a CMO fund and report on a GEO program?
A CMO funds a GEO program in 90 days by making three decisions: name one AI Communications Lead who owns the budget, lock a 12-month budget envelope before execution begins, and commit to a board reporting cadence built on Citation Share. In EPR's experience, programs stall on these executive decisions more often than on execution. The budget ranges and phase splits below are EPR planning estimates, not market benchmarks.
Decision 1: Who owns the budget? GEO sits between earned media, paid acquisition and brand, so without one owner with budget authority it runs as a side project across three functions. Name an AI Communications Lead who reports to the CMO or the CCO, not both. The role is now a real line item in the market, as the Fortune 500 companies hiring GEO managers in 2026 shows.
Decision 2: Build, buy or hybrid? EPR's planning ranges are $400K to $1.5M a year for a mid-market consumer brand, $1.5M to $5M for an enterprise B2B program, and above $5M for a category leader defending a large position. EPR recommends a hybrid: keep measurement and entity work in-house because that capability compounds, and outsource earned media and primary research production. Disqualify any agency that sells GEO as a re-skin of SEO. In regulated categories the bar is higher, as GEO for regulated industries explains.
Decision 3: How does progress reach the board? A program with no board cadence rarely survives its first budget cycle. Frame Citation Share as a pipeline channel, with addressable market, revenue at risk and payback period, on the same dashboard as paid acquisition. Gartner's June 2025 research found that 61% of B2B buyers prefer a rep-free buying experience, and increasingly that research happens inside AI answers. For how to explain SEO and GEO to a board, see the executive briefing section of GEO vs SEO.
| Phase | Days | Executive job | Output |
|---|
| Budget and org | 1 to 30 | Lock the envelope, name the Lead, pick build or buy, brief the board | Approved 12-month budget and org chart |
| Allocation | 31 to 60 | Fund the invisible foundation, track agency burn, set the measurement cadence | Spend matched to the plan |
| Reporting | 61 to 90 | Present day-90 Citation Share to the board and request the next quarter | Three-slide board deck and a funded next quarter |
Days 1 to 30. Lock the envelope first. EPR suggests roughly 5% of the annual budget for diagnosis, 35% for the build phase, 40% for campaigns and 20% for compounding authority work. If the envelope cannot fund all four phases, narrow the program to fewer engines or verticals instead of under-funding all four. A typical team is one Lead, one Content Operations Manager and one Measurement Analyst, plus outsourced earned media and research. Brief the board before execution begins, and use the EPR GEO Scorecard as an external benchmark.
Days 31 to 60. Hold allocation discipline. Sponsored placements in answer engines, where they exist, are not GEO, because they leave no compounding citation share after the spend stops. Fund the invisible work, meaning schema, Wikipedia expansion, Wikidata, Knowledge Graph claims and primary research, since it shows no week-one metrics and carries across years. In a hybrid model, agency spend should match the project plan within about 10%.
Days 61 to 90. The re-audit is the executive deliverable. The board needs the starting position (day-one Citation Share across the four primary engines against the top three competitors), the current position (same method at day 90) and the forward plan. The forward plan matters most, because the board approves the next quarter in the same meeting. The method is in How to Audit Brand Visibility in AI Engines.
| Board slide | Content |
|---|
| 1. Citation Share scorecard | Four engines, twelve queries, three competitors, with the change from day zero to day 90 |
| 2. Revenue impact | Brand share of the category answer surface, addressable revenue at risk where the brand is absent, and payback period |
| 3. Next 90 days | Budget request, headcount request and the two or three Citation Share targets being prioritized |
At day 91 the program becomes an operating function. The quarterly board update recurs, the annual envelope is revisited with the prior quarter's data, and the agency relationship either matures or is re-bid. The execution belongs to the team, and the accountability belongs to the CMO.
For a brand PR team, the cause of AI invisibility is content infrastructure, not the algorithm, and the fix is mechanical. Seven inputs matter, and most teams are thin on at least two.
- Structured site content. Entity-rich, schema-tagged product pages, executive bios, press releases and fact pages, using Article, Person, Organization, Product and FAQPage schema at scale and not only on flagship pages.
- Wikipedia and Wikidata. Accurate, current, well-cited entries for the brand and its top executives, handled through Wikipedia's conflict-of-interest process and never by direct edits from a corporate account.
- Tier-1 earned media. The Wall Street Journal, Bloomberg, Fortune, Forbes and the dominant trade publication in the category.
- Executive digital presence. LinkedIn, bylines, conference appearances and a current, schema-tagged bio on the brand's own site, which engines use to verify expertise.
- Trade and industry research platforms. Statista, the category's dominant trade association and any index the category treats as authoritative.
- Expert-source platforms. ProfNet, Qwoted and Source of Sources, the fastest route to placing a named executive as a quoted source.
- Original data. Surveys, benchmark studies, indexes and white papers, which compound over time instead of decaying like a single press release.
Why do brands go missing from AI answers? Four reasons cover most cases: a thin or inaccurate Wikipedia presence, unstructured website content, absence from secondary sources such as trade publications, analyst reports and expert-source databases, and weak Tier-1 earned media in the past 12 months.
How do PR teams optimize for ChatGPT? They earn presence in the sources ChatGPT draws on and recheck with a fixed prompt set. In EPR's AI Platform Citation Source Index, Wikipedia supplies 26 to 48 percent of ChatGPT's top-10 citations, followed by Forbes, Business Insider, TechRadar, Reuters and PR Newswire. ChatGPT's source mix is the least stable of the five engines, so rerun the prompts monthly. Engine detail is in How to Rank on ChatGPT.
Earned media or brand blog? For most buyer questions engines lean on third-party sources more than a brand's own blog. According to the same index, editorial journalism accounts for about 27 percent of AI citations, rising to 49 percent on time-sensitive queries. Earned media gets the brand named, and owned content supplies the citable facts: original data, definitions, leadership bios and product specifics, tagged with schema. A brand needs both.
What are the five common mistakes? Treating GEO like keyword stuffing, editing Wikipedia directly from a corporate account, handing the work to a junior staffer as a side project, chasing content volume instead of entity clarity, and skipping the baseline measurement.
What is the 120-day build sequence? Days 1 to 30 are the audit: run controlled prompts on all five engines, pull the Wikipedia and Wikidata entries for the company and its top five executives, and score the site for schema completeness. Days 31 to 60 are repair: rebuild executive bios with full schema, update Wikipedia within guidelines, activate executives on expert-source platforms and publish entity-rich pages on topics competitors already get cited for. Days 61 to 120 are compounding: a Tier-1 earned media campaign, an executive byline and conference program, one piece of original research and quarterly Citation Share tracking against named competitors.
Why build before a crisis? A negative news cycle, recall, departure or lawsuit propagates through the AI retrieval layer for months, and a brand has limited ability to remove it. A live library of named executives, recent Tier-1 coverage and original research gives the engines material to surface alongside the crisis coverage, not instead of it. Add Citation Share as a third column next to share of voice and earned media value, as covered in how PR is measured.
Related: The GEO Canon · What Is GEO? · GEO vs SEO · The GEO Operating Stack · The GEO Playbook for Universities · Who Runs AI Communications Inside a Team
