The First 90 Days of Building AI Search Authority

The clock starts the day a brand decides AI search matters. From that point, the next 90 days set the pattern for whether the company becomes a citation source inside ChatGPT, Gemini, and Perplexity or remains invisible to them. AI referral traffic to U.S. retail sites grew 393% in the first quarter of 2026, and AI traffic now converts at 42% better than other sources,. The cost of waiting another quarter is no longer theoretical.

A 90-day plan is not a campaign. It is the period during which a brand earns the right to be cited at all. The phases below are sequential, and each depends on the work of the one before.

Days 1 to 14: Establish the baseline

The first two weeks are spent measuring. Without a starting point, every later result will be guesswork. Teams need to know which prompts customers are likely to ask, how often the brand currently appears in those answers, where the citations are coming from, and how competitors stack up on the same set of questions.

A useful baseline tracks at least 200 to 500 customer-intent prompts across ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews. Each engine treats sources differently. ChatGPT pulls heavily from Wikipedia and Reddit, which together drive a meaningful share of its citations in the U.S. Perplexity leans on primary sources and B2B authority sites. Claude favors The New York Times, The Atlantic, and The Economist.

The baseline output should be one document that names the citation sources currently driving any visibility the brand has, ranks them by authority, and flags the gaps where competitors appear and the brand does not. Without that document, the next two months become guesswork.

Days 15 to 30: Build the infrastructure

The second phase is technical. AI systems read structured data faster than they parse prose, and structured data signals authorship, entity identity, and topic in ways the models trust. Sites that implement complete Tier 1 schema (Organization, Person, FAQ, Product, Article, and HowTo) show measurable visibility gains in AI answers, though the effect is closer to a baseline signal than a single lever/

The research on schema is mixed and worth knowing. Some controlled studies link structured data to citation lift; others have found that pages cited by AI tend to use schema, but the lift itself may reflect underlying content quality rather than the markup alone. The honest read is that schema is necessary infrastructure, not a magic bullet, and FAQ-formatted content tends to perform best for queries phrased as questions, which now describe most inputs to conversational search.

Other infrastructure work belongs to the same window. Author bios should be rebuilt with sourceable credentials and links to verifiable external profiles. Organization pages need consistent NAP data and accurate financial or executive information. Older content with outdated facts should be corrected, since AI systems retrieve and weight pages by perceived accuracy.

The work is unglamorous and easy to skip. Skipping it is the single most common reason 90-day GEO programs fail.

Days 31 to 60: Earn citations from the sources that matter

The middle month is when authority is built. AI engines cite the open web they trust, and the open web they trust is overwhelmingly earned media, not owned media. Generative Pulse research from May 2026 found that earned media accounts for roughly 84% of all AI citations, with journalism alone making up 27% of cited links.

The implication is direct: the brand blog is not the lever. The Wikipedia entry is. The Reddit thread is. The Forbes column, the trade journal interview, the YouTube explainer, and the analyst quote are. Reddit is now the single most-cited domain across all major AI engines.

Activity during this window should concentrate on three fronts. First, verifying or expanding any Wikipedia presence the brand has, with attention to neutral sourcing standards. Second, securing earned coverage in the editorial outlets each target AI engine actually trusts. ChatGPT favors Wikipedia and large editorial publications, Perplexity rewards primary research and trade publications, and Claude leans toward long-form journalism in The New York Times, The Atlantic, and The Economist. One PR push will not move every engine equally. Third, cultivating participation in the community platforms where the brand's category gets discussed, with disclosure and care. Reddit and YouTube together now account for a large share of citations across every major engine, and ignoring them concedes ground to competitors willing to do the work.

Press releases, despite the volume of them produced each year, account for only about 1% of AI citations across major engines. A 90-day plan that relies on them is built on a false premise.

Days 61 to 90: Measure, iterate, and prove the model

The final 30 days are about confirming whether the work has moved the numbers and deciding what to scale. Reports run in this window should compare against the Day 1 baseline on the same prompt set. Established brands running comprehensive programs typically see meaningful AI visibility gains by Day 90, with initial improvements often visible within the first month to 60 days.

Two patterns tend to emerge. Schema and content infrastructure changes show up first, often within four to six weeks of implementation. Citation gains from earned media take longer to register because AI models refresh their retrieval corpora on different cadences. Citation share for any single source can shift dramatically week to week. ChatGPT's Reddit citation share fell from roughly 60% to 10% in about two weeks in September 2025, after Google changed its indexing in a way that made Reddit content harder for LLMs to crawl. Brands that diversify their source profile weather these shifts better than those concentrated on one channel.

The reporting at Day 90 should show citation rate, share of voice, sentiment, and source quality for each engine, with the baseline numbers next to current ones. A program that has moved one of the four meaningfully has earned the right to a second quarter. A program that has moved none of them has a diagnostic problem to resolve before more spend follows.

What success actually looks like

After 90 days, a well-executed AI search authority program will not yet dominate any model. It will have proof points: measurable citation lift on a defined prompt set, schema and content infrastructure in place, earned coverage in two or three engine-relevant outlets, and a sentiment baseline against which to measure year two. That is enough to keep going. Compare that to a brand still arguing about whether AI search is real: the same 90 days will have produced no baseline, no schema, no citations worth tracking, and a competitive position harder to recover.