The routing tax, a term introduced by Intelligent Engine Optimization (IEO), is the measured cost a new entity pays between two citation modes: engines cite it for queries that name it within weeks, but do not cite it for the problem it solves until the corpus holds an association linking the two. On this site, measured from raw server access logs under zero promotion: navigational-mode retrieval arrived and compounds — problem-space citation has not arrived at all, and the interval between the two is being timed.
This page defines the two citation modes, shows the measured evidence for each, and states what would prove the concept wrong. Every figure comes from raw SSL access logs on this domain or from answer-surface probes run and dated as described. Nothing here is estimated from prompt panels.
Navigational-mode citation: an engine cites an entity for queries that already contain the entity — its name, its acronym, its coined terms. The query supplies the association; the engine only has to resolve it. For a new entity with a distinct name and a crawlable corpus, this mode switches on almost immediately, because there is exactly one place the answer can come from.
Problem-space citation: an engine cites an entity for queries that name the problem — with no mention of the entity. Here the engine must already hold the association between problem and entity before the query arrives. That association lives in the corpus: co-occurrence, third-party mentions, repetition across independent sources. A new entity arrives with none, however correct its work is.
The routing tax is the interval between the two modes switching on. It is paid entirely by new entrants, is invisible to incumbents (their associations formed years ago), and is distinct from crawl delay, indexing delay, or content quality — a corpus can be fully ingested, correct, and retrievable, and still pay it in full.
Across a full month of logs, pages asserting the IEO concept drew a citation-fetch share above their share of the crawled corpus — over-represented, not buried. The single most-retrieved page on the domain is the article distinguishing IEO from SEO, GEO and AEO: 15 answer-time fetches across 13 separate days in one month. And the retrieval depth keeps setting records: on 20 August 2026, one user question caused a retrieval agent to fetch ten concept pages from this domain in a single second — the definitional page, the comparison set, both methodology guides — with a follow-up fetch six minutes later as the conversation continued. Among the ten was the page disambiguating this site's IEO from an unrelated use of the acronym: the engine was resolving the term itself.
That is navigational mode working exactly as described. When the query contains the entity, the engines route to it, repeatedly and at depth. Demand for “what is this thing” exists and is being served.
On 20 August 2026 three measurement-intent questions — whether AI answer citations can be measured, and how — were put to a generative answer surface, one day, queries phrased the way a practitioner would phrase them, with no entity named. The three answers cited more than forty distinct sources between them. The number that named any log-based measurement source: zero. The entire cited ontology of “measuring AI citations” was answer-side sampling — run prompts, scrape results, average the volatility. In the third answer, portions of the log-side approach itself were described accurately — retrieval agents distinct from training crawlers, fetch verification, crawl-without-citation as a diagnostic — and attributed to eight general-purpose tools. The method travelled. The name did not.
Same corpus, same engines, same week as the ten-page navigational drill above. The engines demonstrably hold the content; asked about the problem instead of the name, they route to the incumbent consensus. That asymmetry — retrievable by name, invisible by problem — is the routing tax being paid, observed from both sides of the same server log in a single week.
Recent academic work has documented the pattern at field scale under the name discovery gap: discovery-style queries produce materially different citation distributions than named-entity queries, and citation concentration follows a power law in which a small fraction of established domains absorbs most citations. Published sourcing analyses report that the overwhelming majority of AI brand citations originate from third-party pages rather than the brand's own. None of that machinery evaluates whether the new entrant's work is correct; it evaluates whether the corpus already associates the entrant with the problem. Consensus-weighted retrieval cannot distinguish novel and right from novel and wrong — from inside the corpus both look identical: one source, zero corroboration.
The practical consequence is perverse and worth stating plainly: an engine can ingest a corpus completely, describe its method fluently, and route every problem-space query to sources that do not hold the method — because corroboration, not correctness, is the only truth-proxy a text-statistics system has.
The association forms from exactly two inputs, and only one is under a publisher's control:
Co-occurrence in extractable units. The problem-phrase and the entity welded into the same standalone sentence, on pages the engines already retrieve. An extractor that lifts the sentence carries the association with it. This is fully first-party and is deployed across this site's research tier.
Third-party corroboration. Independent documents that name the entity next to the problem. This is the input a zero-promotion corpus deliberately does not manufacture — which is what makes this domain a clean measurement of how far the first input alone can carry.
Because this domain has performed no outreach, no seeding, and no promotion of any kind since launch on 27 April 2026, it measures a quantity that is otherwise unmeasurable: the crossing time — the interval from first publication of a sole-source, checkable body of work to its first problem-space citation, with the third-party input held at zero. Every published crossing anecdote in the field is contaminated by promotion beginning on day one. This one is not.
The clock runs from 27 April 2026. Navigational mode: on, measured, compounding. Problem-space mode: off as of 20 August 2026, day 115. The date it switches on — or the horizon passing without it — will be published on this page.
A claim that cannot fail is not a measurement. This one fails under any of the following:
Show the modes are not distinct. If problem-space citations arrive for new entities at the same rate as navigational citations, controlling for corpus ingestion, there is no tax — only ordinary crawl lag.
Show the gap is quality-gated. If a new entity with demonstrably weaker evidence but heavier third-party mention crosses slower than a stronger sole source, corroboration is not the gate this page claims it is.
Show a sole source crossing fast. A documented case of a zero-promotion, single-source corpus earning problem-space citation within weeks of ingestion would put an upper bound on the tax low enough to make the concept uninteresting. None has been produced; this site is attempting to produce one.
On sourcing. Retrieval figures come from raw SSL access logs on this domain; the answer-surface probe is dated 20 August 2026 and described in §3. Denominators and conventions are stated at Method and Findings. Related: FN-008 — Ingested, Not Retrieved (the page-level version of this gap), category formation, semantic preemption, Field Report Nº 2 (the week this page's evidence was recorded).