The short version
TSM on a Napkin
The whole Three Streams GEO Methodology on one page, in about five minutes. Each part links to the full explanation when you want more.
1
Someone asks an AI. It names a brand. Is it yours?
More and more people ask an AI instead of searching, to buy, learn or check a fact. It writes one answer and cites a few pages. There is no page two.
GEO is the work of being in that answer, and being described accurately. TSM is a method for doing it, with three teams pulling together.
2
How an AI answer happens
Swipe for the next step →
-
0 Before any prompt Crawlers store your page in the index, cut into passages.
-
1 Someone sends a prompt: a question, a problem or a product search.
-
2 The engine writes its own searches: the fan-out queries.
-
3 Each search is matched against the stored passages, by meaning and words.
-
4 The results are merged and ranked. A few pages are picked.
-
5 A live reader opens only those pages and reads their HTML, not JavaScript.
-
6 It picks passages, writes the answer and cites some sources.
-
↺ Thinking modes may loop from step 5 back to step 2 and search again.
Google’s AI Overviews and AI Mode have no documented live reader; they answer from Google’s index.
3
Three streams. Miss one, lose the answer.
Each stream makes one thing true. All three must hold at once.
- Technical: found and readableCrawlers can index every page, and key facts are in the server HTML.Without it: citations go to other sites’ pages.
- Content: on topic, worth quotingTitles and paragraphs match the fan-out queries; passages hold specifics.Without it: no page of yours is cited.
- Business: known and trustedReviews, press, communities and Wikidata teach the models who you are.Without it: a better-known rival is named instead.
4
A program moves in phases. Gates decide, not the calendar.
- 0Capability BuildingAssemble the team, train it, set up measurement, record a baseline.Exit when: the team is trained and the baseline is recorded.
- GateGO?
- 1FoundationCore technical work, small pilots to prove the model.Exit when: first AI citations are documented and measurement is validated.
- GateGO?
- 2ScalingExpand what worked, publish original research, build authority.Exit when: how often you are cited rises beyond normal run-to-run swings, and you gain on competitors.
- GateGO?
- 3Excellence & SustainabilityOptimize from data, write the playbooks, run it as operations.Exit when: you reach the position you set as your goal, and the processes are written down.
At each gate the evidence decides: GO, EXTEND, ADJUST or PAUSE. The phases and their success criteria
5
What a plan looks like
Each step builds on the one before, and the streams hand work to each other. Measuring starts the next round.
Business
Content
Technical
- 1Pick a heroThe product, service or cause to start with
- 2Write sentinel promptsWhat people ask about it, without your nameUses 1
- 3Lock the competitor setThe names that show up for those promptsUses 2
- 4Plan and write piecesEach with a clear job; must pass the quality gatesUses 1, 2 and 3
- 5Publish and get indexedLive, crawled, in the indexUses 4
- +Alongside: earn mentionsReviews, press and communities, so the models know you
- +Always on: crawlers in, facts in the HTMLAccess for search crawlers, server-rendered pages
- 6MeasureSame prompts, many runs, against your baseline and competitors. The result picks what comes next.Uses 2, 3 and 5; then back to 1–4
6
How you know it’s working
AI answers change every time, so one check proves nothing. TSM sends the same sentinel prompts many times, on each engine, and reads three things:
- Seen: how often engines cite your pages and name you, next to your competitors.
- Right: whether what is said about you is accurate.
- Worth it: branded searches and visits that come from AI, and what they lead to.
No targets: change is read against your own starting point.
7