We Benchmarked the Websites of 6 of America's Largest Independent Oil Refiners

SummaryWhat 6 oil refining sites reveal

Of the 6 sites we measured, 4 emit zero structured data (no JSON-LD blocks at all) and 1 ship no meta description on the page we tested. Those two fields are exactly what an AI answer engine reads first when it decides whether - and how - to cite you. A page can dominate Google for its brand name on the strength of decades of links and still be nearly invisible to a model that is summarizing the oil refining category from scratch, because the model has nothing clean to extract.

01Method & dataset

We picked the 6 oil refining companies below and ran each homepage through Harvv's free audit - the same one any visitor can run at harvv.com. Each site was loaded multiple times in a real headless browser at a phone viewport (390px wide) and a desktop viewport (1366px), and scored with Google Lighthouse. We recorded load speed (Largest Contentful Paint), layout stability, tap-target sizing, the SEO fundamentals on the page (title, meta description, canonical, H1), and the structured-data blocks (JSON-LD) present in the HTML.

Every number on this page comes straight from those audits. Where a cell reads "n/a" the audit did not return that metric. This is a measurement of public homepages on the dates shown, not a controlled experiment, and homepages are only one page of a large site - see limitations.

02The AI-readiness gap is the real story

Search is splitting in two. The classic path - type a query, scan ten links, click - still exists, and on that path these companies win, because authority compounds over decades. The new path is an answer: a model reads the web and hands the user a synthesized paragraph, sometimes with citations, often without a click. On that path the rules invert. The model is not weighing your backlinks. It is parsing your markup.

4 of 6 emit no structured data at all; the 2 that do expose only WebPage. For a category leader, that is the equivalent of running a flagship store with no signage a machine can read. 1 of 6 also leave the meta description blank, so when a model (or Google) needs a one-line summary, it scrapes whatever text it finds first - rarely the pitch you would choose.

None of this is a speed problem or a design problem. These are some of the most trusted names in oil refining. It is a machine-legibility problem, and it is invisible in every tool that only looks at rankings - which is precisely why it persists.

03The benchmark

Every site, head to head. LCP is median mobile load; Perf / SEO are Lighthouse scores (0-100); Tap % is the share of undersized touch targets on mobile; Meta is whether a meta description is present; JSON-LD is the count of structured-data blocks.

Median across the set: mobile LCP 0.61s, Lighthouse performance 56/100 mobile and 76/100 desktop, SEO 85/100. 2 of 6 ship no canonical tag; 2 of 6 do not have exactly one H1 on the page we tested.

04Speed and mobile: better than you'd guess, with sharp exceptions

Raw speed is the one area where pedigree shows. Valero painted its main content in about 0.43s at the phone viewport in our loads - comfortably inside Google's 2.5s "good" band. The slowest in the set, HF Sinclair, took about 1.02s. The median, 0.61s, is genuinely respectable.

Mobile usability is where the catalogs show their age. The median site has 68.5% of its tappable elements under the 44px touch-target floor - dense link lists and tiny icons built for a mouse, not a thumb. On 2 of 6 sites, effectively every image on the homepage ships without width and height set, so the page reflows as it loads and a visitor reaching for one link lands on another.

05What to fix first

In priority order, for any site in this set - or any oil refining site with the same profile:

1. Ship Organization + Product/Service JSON-LD. This is the single highest-leverage change for the AI era and the one 4 of these 6 sites skip entirely. It is the markup an answer engine reads to know who you are and what you sell.
2. Write a real meta description on every important page. 1 of 6 leave it blank, handing the summary to a scraper.
3. Set width and height on every image. It stops the layout from jumping as the page loads - the most common cause of mis-taps on these sites.
4. Grow the touch targets. The 44px floor is not cosmetic; on a thumb, undersized links are the difference between a tap and a miss.
5. Add a canonical tag (2 of 6 have none) so parameterized URLs don't split ranking signal.

06The individual teardowns

Each company's full report - every finding, on phone and desktop, with the fix:

• Valero → full teardown(3 findings)
• Phillips 66 → full teardown(3 findings)
• HF Sinclair → full teardown(5 findings)
• PBF Energy → full teardown(3 findings)
• Par Pacific → full teardown(5 findings)
• Delek US → full teardown(6 findings)

07Limitations

This is an honest measurement, not a verdict on these companies' businesses. Caveats: (1) We tested homepages only - large catalogs may carry structured data and meta on deeper product pages we did not crawl. (2) Lab loads from a single location are not field data from real users on real networks; treat the speed numbers as relative, not absolute. (3) Lighthouse scores vary run to run; we report medians but a few points of wobble is normal. (4) "Zero JSON-LD on the homepage" means exactly that - it is not a claim about the entire site. The pattern across the set is the signal; any single cell deserves a second look before you act on it, which is what the individual teardowns are for.