Document Intent: When Buyers Search the Report, Not the Topic

A thirty year old retrieval category, newly visible in energy

The behaviour described here is not new and does not need a new name. Library and information science has called it known item search since at least Wildemuth and O'Neill's 1995 work, and Lee, Renear and Smith formalised the variations of the concept in a 2006 ASIST paper. The defining feature is that the searcher is not exploring. They already know the thing exists, they know roughly what it is called, and the search is an act of retrieval rather than discovery.

Andrei Broder carried the idea into web search with his taxonomy of web search, which split queries into navigational, informational and transactional. His own description of the navigational case is the clearest statement of the problem a third party faces: the purpose of such queries is to reach a particular site that the user has in mind. Broder measured navigational queries at 20 per cent of a query log and 24.5 per cent of a user survey. Those figures come from AltaVista era data and should be treated as conceptual scale rather than a current measurement, but the category has never gone away.

What has changed is the density of named documents in professional markets. An energy buyer in 2026 is surrounded by titled artefacts that carry real commercial weight. BP published Growing shareholder value: a reset bp on 26 February 2025. Shell ran a Capital Markets Day in 2025. OPEC launched the World Oil Outlook 2025 on 10 July 2025. The IEA publishes the World Energy Outlook annually. The European Union's CBAM regulation and the CSRD directive have formal numbers that people search. DeGolyer and MacNaughton is conducting an OPEC capacity audit that will feed 2027 baselines, and people are already searching for the report by name.

Each of those is a thing a professional can name. When they want it, they do not search for what is in BP's strategy. They search for the title, and often they add pdf. That is document intent, and it is a distinct demand shape from the question led content most energy B2B programmes are built to serve.

المشروع 54A backlit keyboard in a darkened room. Document intent search is a typing behaviour before it is a marketing problem: a professional who already knows the title and types it exactly.

Measurable impressions, zero clicks, and a page routing fault

Project 54 publishes research for energy B2B and has three years of search data for a site that writes about exactly these documents. The 90 days to 9 October 2026 produced 32,978 impressions and 107 clicks sitewide, a 0.32 per cent click through rate. Inside that, the document intent pattern is clearly visible.

The strongest single cluster is built on the exact title of BP's February 2025 strategy document. The bare title, the title with 2.3 appended, the title with 2.3 and 2.5, the title with transition businesses, and the title with pdf together drew roughly 574 impressions at average positions ranging from 5.6 to 20.1. Clicks across the entire family were zero. A second cluster, shell global net-zero strategy, drew 602 impressions at average position 9.0, again with zero clicks, and shell global corporate sustainability reporting drew a further 245. A query for a named consultancy report, the DeGolyer and MacNaughton OPEC 2027 capacity review, appeared at average position around 5.

The most instructive finding is not the volume. It is which page Google chose. For the Shell net zero queries, the URL being shown was a buyer journey statistics article, not the site's dedicated Shell net zero strategy page, which exists and did not rank. That is a relevance and internal linking fault inside one site, not a statement about demand. It means a share of these impressions reflect Google's uncertainty about which page answers a titled query, rather than genuine consideration of a candidate result.

The honest summary of the data is therefore two sided. Document intent demand exists, it is sizeable relative to this site's total impressions, and it sits at positions where a better result could plausibly win attention. It has also produced no clicks at all, and at least one large slice of it was being served by the wrong page.

PDFs rank badly, and the obvious response is the one Google penalises

The first reason is mechanical. Google does index PDFs, but its own documentation on indexable file types places them among formats that are binary files or complex containers that require a specific parser to extract the human readable text. Detection depends on the server sending the right Content-Type header. A PDF is not a first class document to a search engine, it is a parsing job. That is why the official document frequently does not dominate searches for its own title, and why positions five to twenty are available at all.

The second reason is that the obvious commercial response is close to the behaviour search engines have spent two decades suppressing. If you publish a page whose purpose is to rank for someone else's document title and whose content is a summary of that document, you are squarely inside Google's own warning questions. Its helpful content guidance asks whether the content provides substantial value when compared to other pages in search results, and whether you are mainly summarising what others have to say without adding much value. Its spam policies name republishing content from other sites without adding any original content or value. In November 2025 Pandu Nayak, Chief Scientist for Search, described parasite SEO as a pay to play scheme designed to fool our ranking systems and users.

None of that forbids writing about a public document. It does mean the version of this tactic that is easiest to execute at volume is the version most likely to be devalued, and that anyone proposing a document intent programme should be asked what their page does that the PDF does not.

The third reason is that the surface itself is shifting. Semrush's AI Overviews study found that the share of navigational queries triggering an AI Overview rose from 0.74 per cent in January 2025 to 10.33 per cent in October 2025, a roughly fourteen fold increase in ten months. Even where a titled query once produced a clean list of links, it increasingly produces a synthesised answer first.

The counter argument, stated properly

A reader who stops here should be able to argue the opposite position, so here it is at full strength.

Known item search is defined by the fact that only the item satisfies it. Broder's own framing predicts that a third party page is an obstacle rather than a destination. Project 54's data is consistent with that prediction and not with the opportunity reading: 574 impressions and 602 impressions produced zero clicks between them. If the demand were capturable from position nine, something would have been captured.

Impression volume is also a weak proxy for available traffic in general. The SparkToro and Datos zero click study put US zero click searches at 58.5 per cent and European at 59.7 per cent, with roughly 360 clicks to the open web per 1,000 searches. A thousand impressions on any query is not a thousand opportunities.

There is the quality objection already made, and a measurement objection on top of it. We could not find a single published case study of a publisher deliberately ranking for a primary document's title, and therefore no evidence of any kind about whether such traffic converts. In the absence of conversion evidence, impressions on document titles are a vanity metric, and an energy supplier has better uses for a content budget.

Finally, some of the impression volume in this dataset is probably not human. Two of the largest zero click queries on the site, eni business model b2b at 5,610 impressions and average position 5.0, and an asset acquisition framework query at 9,073 impressions and average position 1.9, behave like machine generated probes. A human query at position 1.9 produces clicks. Several other long natural language queries on the site read like prompts typed into an assistant rather than a search box. Any sizing exercise built on raw impressions will overstate the prize.

Fix routing, compete on extraction, and instrument before scaling

Start with cannibalisation, because it costs nothing and the evidence for it is unambiguous. On this site, five separate URLs compete for the query eni business model b2b, which draws 5,610 impressions at average position 5.0 and earns no clicks at all. A Shell query is being answered by a buyer journey statistics page while the real Shell page sits unranked. Neither of those is a content gap. Both are internal relevance faults, and both are cheaper to fix than to write around. Audit which of your own URLs Google associates with each titled query before you commission anything.

Second, if you do publish against a named document, compete on the job the PDF cannot do. The defensible asset is extraction: the capex table, the segment definitions, the Scope 3 boundary change, the year on year deltas, in HTML, in a table, sortable, with the figures stated plainly. That passes Google's own test of providing substantial value compared with other pages in search results, and it is the thing a reader who has the PDF open still cannot get quickly. Link out to the official document above the fold rather than hiding it. A reader who leaves for the primary source and remembers who sent them there is a better outcome than a reader who bounces.

Third, build the page as the navigational query it is. The exact document title belongs in the H1 and the title tag. Ahrefs' analysis of 1.4 million prompts found that pages with natural language URL slugs were cited by ChatGPT at 89.78 per cent against 81.11 per cent for others, so the slug should carry the title too. Add FAQ entries for the modifiers people actually append, which in this dataset were pdf, a section number, and a named chapter.

Fourth, instrument before you scale. Nothing in the literature tells you whether this traffic converts, so treat it as an experiment with a stopping rule. Build five to ten companion pages, only for documents where you have genuine proprietary commentary, and measure assisted conversions and known account visits rather than impressions. If the pages become the clear best result and clicks still do not come, the counter case has won and the programme should stop. That is a perfectly good outcome, and it is cheaper than discovering it at scale.

The wider point for energy marketers is about where demand actually lives. The same SparkToro and Datos work found 44 per cent of searches are branded and the true long tail is 3.6 per cent of volume. Search demand concentrates on names: company names, product names, and document names. A content programme built entirely around questions is built around the smallest part of the market.