AI Mode vs Calculator Publishers: 10,000 Traffic-Mix Scenarios

Search features can change how users discover finance content. This page tests one explicit scenario model: results depend on the exposure and CTR assumptions you choose. We modeled 10,000 synthetic publishers to show the range, not to predict Google's ranking or any site's future traffic.

Finding 1: The modeled category multiplier drives the result

Per-query-type loss multipliers (AI exposure × CTR drop when shown):

Under these assumptions, the tool-required category loses 6%. Definition queries lose 53%. A publisher's traffic loss is is driven mainly by the share of traffic in each bucket. The simulation does not test content quality, age, domain authority, schema or Google's ranking systems.

Finding 2: This simulation produces two clusters

The synthetic 10,000-publisher loss distribution doesn't form a smooth bell curve. It bunches around two centers — one near 24.7% (tool-dominant mixes) and one near 39.4% (definition-dominant mixes). A publisher can use its own query data to test how sensitive the result is to the mix; this model cannot predict an individual site's fate.

Finding 3: Test page structure before commissioning new content

A publisher with a 60% definition-content / 40% tool-content mix loses ~34% of traffic. The same publisher with a 40% definition / 60% tool mix loses ~24% — roughly half in this model. A practical experiment is to make the relevant calculator easier to find on pages where it genuinely answers the query; add `SoftwareApplication` schema only when it accurately describes the visible tool. Schema should document the page, not try to force a category.

Finding 4: Make provenance and answers easy to evaluate

Search features can cite pages in different ways. The following are practical patterns to test, not guarantees of inclusion in AI Overviews:

• Specific numerical claims with assumptions and primary-source citations — state the rate, timing, fees and source; a calculator output is an illustration, not a forecast.
• Dataset / Original-Research schema — use it when the page publishes reproducible data; it clarifies provenance but does not guarantee a citation.
• Tight, structured answer paragraphs at the top of the page — what we call "speakable summary" — formatted so the AI can lift one block as a citation.
• Clear author and publisher identity — use Organization/Person references only when they are accurate and verifiable; do not add social or directory links as a ranking tactic.

A defensible workflow is: (1) add a tool only when it solves the page's query, (2) publish reproducible research and label it accurately as `Dataset`, and (3) keep author and organization schema accurate and tied to the actual publisher.

Sample publishers (8 across the loss spectrum)

Methodology

• 10,000 publishers, deterministic via Mulberry32 PRNG seeded 20260524.
• Traffic mix per publisher: Dirichlet-style random share across 5 query types, normalized to sum to 1.
• Baseline traffic: uniform 1,000–1,000,000 monthly visits.
• 5 query types with these modeled parameters (AI exposure × CTR drop when shown):
• Definition / explainer ('what is compound interest'): 85% × 62% = 53% net loss.
• Tutorial / how-to ('how to calculate APR'): 78% × 55% = 43% net loss.
• Comparison ('snowball vs avalanche'): 70% × 48% = 34% net loss.
• Best / review ('best HYSA 2026'): 68% × 45% = 31% net loss.
• Tool-required ('compound interest calculator'): 22% × 25% = 6% net loss.
• Post-AI traffic per publisher = Σᵢ (mix[i] × baseline × (1 − exposureᵢ × dropᵢ)).

Where the parameters come from

AI exposure and CTR-drop values are scenario inputs, selected to make sensitivity visible across five query categories. They are not measurements of Google results or of any named publisher. Replace them with a verified dataset when you have one, then rerun the calculation to test the effect.

These are modeled estimates, not measurements of any specific publisher's traffic. The simulation's value is the relative spread between query-type winners and losers — that spread can change when the assumed exposure or CTR values change; use the sensitivity controls and report the assumptions alongside any reuse of these numbers.

Limitations

• 5 query types is a simplification. Real publishers have dozens of subcategories; some (commercial-investigation, navigational) we omitted because their AI Overview behavior is still settling.
• Revenue ≠ traffic. CPM and conversion rates differ by query type, but this simulation does not model monetization or revenue resilience.
• The model is a one-step scenario. Search features, user behavior and CTR can change over time; rerun with new observations instead of treating one output as a forecast.
• Citation-rate-back-to-publisher is not modeled. A cited page may receive traffic, but the size and direction of that effect are not assumed here.
• Direct + referral + email traffic are unaffected by AI Overviews and not modeled in the simulation.