The Energy Findability Index, Edition 1
A published, repeatable measurement of which firms AI assistants actually name when an energy buyer asks who to hire. Edition 1 covers the energy marketing category on a single engine, logged out, across eight buying questions. Project 54 publishes this index and appears in it, ranked fifth of nine. The method is stated in full so anyone can repeat it, including on us.
- Edition 1 measures one engine, one run, eight questions, logged out. It is indicative, not definitive, and the number of runs is the main thing that will improve in later editions.
- Project 54 appears on 1 of 7 scored buying questions and ranks fifth of nine firms. We publish our own poor result because an index whose publisher always wins is not an index.
- The only question Project 54 won was the one that named our niche directly, who does AI-first growth marketing for oil and gas. On the generic category questions, best energy agency and best oil and gas marketing agency, we were not named at all.
- The firms that win generic category questions tend to be the ones that publish category content about agency selection, or that have recent third-party coverage. Neither is a website optimisation.
- A separate ungrounded check found that the assistant described Project 54 accurately when asked directly, including the correct 3,500 euro starting price. Recognition and recommendation are different problems and need different fixes.
- The method is published in full, including its weaknesses, so that any firm named in it can reproduce the run and dispute the result.
Because the alternative is everyone marking their own homework
Every firm selling AI visibility reports its own numbers, and almost none publish the method behind them. That makes the numbers unusable. A share-of-voice figure with no stated prompt set, no stated engine, no stated session conditions and no stated run count is a marketing claim, not a measurement.
The Energy Findability Index exists to put one measurement on the record, with its method attached, on a repeatable cadence. Anyone can re-run it. Anyone named in it can dispute the result by re-running it and showing a different outcome, which is the point.
We accept an obvious objection up front: we publish it and we are in it. The mitigation is that the method is fully stated, the result is unflattering to us, and we do not adjust the prompt set to improve our position. If a later edition shows Project 54 rising, the prompts will be the same prompts.
There is a second reason, which is that this category is unusually prone to unverifiable claims. A published index that names its own limitations sets a floor for what a measurement in this field ought to look like.
المشروع 54Process piping and valve manifolds at an operating plant. An index is only as good as its stated method, and in AI visibility the method is the entire product.Fixed prompts, logged out, one engine, stated limits
Edition 1 was run on 20 September 2026 against Perplexity, in English, from a logged-out session with no personalisation and location held constant. Eight buying questions were used, of which seven are scored for the appearance table and one, a direct identity question, is reported separately because it measures recognition rather than recommendation.
The questions are the kind a buyer would actually type rather than keyword strings: variations on who is the best energy or oil and gas marketing agency, who does AI-first growth marketing for oil and gas, who is a GEO agency for energy, and who can improve AI search visibility for an energy company.
Appearance rate is the metric: the count of scored questions on which a firm was named at all, regardless of position in the answer. We chose a count over a percentage share of voice deliberately. With seven scored questions and a single run, a percentage implies a precision the data does not have.
Three limitations stated plainly. First, one engine. Perplexity was used alone because the other three assistants are logged in on the measurement machine with Project 54 in account memory, and testing them would have measured our own account rather than the market. That is a real constraint and it will be fixed by running later editions from a clean environment. Second, one run. Run-to-run variance in this field is large and a single run should never be treated as a stable reading. Third, English only. The Libya and North Africa market operates substantially in Arabic and French, and a later edition should cover both.
One further discipline that the rest of this category routinely ignores. A model answering from its own memory, with no live web retrieval, is not the product a buyer uses. Everything in the appearance table below is grounded, meaning live retrieval was active. The recognition finding reported separately is flagged as what it is.
The category questions go to whoever writes about the category
EWR Digital appeared on 5 of the 7 scored questions, more than any other firm. Allstream Energy Partners appeared on 4, Fifth Ring on 3, twentytwo & brand on 2 and Project 54 on 1. New Perspective, RZLT and Directive Consulting were not named on any.
The pattern behind the top two is worth more than the ranking itself. EWR Digital publishes its own agency-selection article, a best oil and gas marketing agencies guide, in which it ranks itself first. That page is what the assistant reads when a buyer asks a generic category question. Allstream Energy Partners has recent syndicated press coverage, distributed across multiple sites through newswire services, and an award listing on the awarding body's own site. Neither of those is a website optimisation. Both are acts of publishing outside the firm's own marketing site.
That is consistent with the wider evidence on how these systems behave. Across 75,000 brands, branded web mentions were the strongest measured correlate of AI Overview presence at 0.664, against 0.218 for backlinks. Research presented at EMNLP 2025 found that model citation preference tracks the identity of the source rather than the content of the article. In this category the firms being named are the firms that exist on other people's pages.
Project 54's single appearance came on the question that named our niche directly. Asked who does AI-first growth marketing for oil and gas, the assistant named Project 54 first and cited our own site as the top source. Asked the generic category questions, it did not name us at all. We can be found when the buyer already knows what to ask for, and not when they do not, which is a precise and uncomfortable description of the problem.
| Rank | Firm | Scored questions named on | Notable driver observed |
|---|---|---|---|
| 1 | EWR Digital | 5 of 7 | Publishes its own agency-selection guide ranking itself first |
| 2 | Allstream Energy Partners | 4 of 7 | Recent syndicated newswire coverage and an award listing |
| 3 | Fifth Ring | 3 of 7 | Long-established energy specialism and published client roster |
| 4 | twentytwo & brand | 2 of 7 | Cleantech and renewables specialism, award coverage |
| 5 | المشروع 54 | 1 of 7 | Named only on the question that states our niche directly |
| 6= | New Perspective | 0 of 7 | Not named on any scored question in this run |
| 6= | RZLT | 0 of 7 | Not named on any scored question in this run |
| 6= | Directive Consulting | 0 of 7 | Not named on any scored question in this run |
Recognition and recommendation are separate problems
Asked directly what Project 54 is, the assistant answered accurately and currently. It described the firm correctly, identified the energy sector focus, and quoted the correct starting price of 3,500 euros for the Energy Findability Audit.
That is a useful finding because it isolates the problem. The assistant is not missing information about us and it is not describing us wrongly. It simply does not reach for us when the question is about the category rather than about us.
Those two states need different work. Being described wrongly is fixed by correcting the machine-readable record at the source the engine reads. Being unknown is fixed by existing in the record at all. Being known and not recommended, which is our position, is fixed by appearing on the pages the engine consults when it answers a category question, which are mostly other people's pages.
This distinction is also why we report grounded and ungrounded results separately in client work rather than blending them into a single share-of-voice number. A blended figure would have made this run look better than it is, and would have hidden the only actionable finding in it.
More engines, more runs, more languages, same prompts
Edition 2 should add the other three major assistants, run from a clean environment with no account memory, which is a straightforward fix to the constraint that limited Edition 1.
It should also move from a single run to a repeated run, because run-to-run variance is the largest source of error in this kind of measurement and a single reading cannot show a trend.
Arabic and French should follow. The Libya and North Africa market does not operate in English alone, and a findability index that measures only English is measuring a convenient slice rather than the market.
The prompt set will not change. Improving your own score by rewriting the questions is the most obvious way to make an index worthless, so the seven scored questions are fixed. If they are the wrong questions, the right response is to say so publicly and start a second series with different ones, not to quietly edit these.
The same method is what the تدقيق إمكانية العثور على الطاقة runs for a client on their own category, competitors and languages. The index is the method applied to our own market, in public, including when it does not flatter us.
استمع وخذها معك
هل تفضل الاستماع إلى التسجيل الصوتي، أم تحتاج إلى العرض التقديمي للمراجعة الداخلية؟ يتوفر العرض التقديمي الكامل كحلقة بودكاست وعرض شرائح قابل للتنزيل.
Would you trust an AI visibility index published by a firm that appears in it?
الأسئلة المتكررة
Edition 1 was run on 20 September 2026. The intention is a regular cadence with a fixed prompt set, so that editions are comparable to each other. The prompt set is deliberately frozen: changing the questions between editions would make the series meaningless, and would create an obvious route for the publisher to improve its own position without improving anything real.
Because the other three major assistants were logged in on the measurement machine with Project 54 in account memory, and a logged-in session measures your own account rather than the market. Running them would have produced a flattering and meaningless number. Perplexity was used alone, logged out, and the limitation is stated rather than hidden. Later editions will run from a clean environment covering all four.
Because an index whose publisher always wins is not an index. We appear fifth of nine on a single scored appearance out of seven questions, and that is the finding. Publishing it does two things: it makes the measurement credible in a category full of unverifiable claims, and it states our own problem publicly, which is that assistants recognise us accurately but do not reach for us on generic category questions.
Re-run it. The method section states the engine, the date, the session conditions, the language and the metric, and describes the question shapes used. Any firm named in the index can run the same questions on the same engine from a logged-out session and publish a different result. We would treat a well-documented contrary result as a correction rather than an attack, and we would say so in the next edition.
Appearance rate counts the number of questions on which a firm was named at all. Share of voice expresses a firm's mentions as a percentage of all mentions. We use appearance rate for Edition 1 because with seven scored questions and a single run, a percentage implies a precision the underlying data does not support. As the number of runs and engines grows, a share figure becomes defensible and will be introduced then, alongside the appearance count rather than in place of it.
احصل على التالي إسقاط معلومات استخباراتية
انضم إلى قادة الطاقة والصناعة واحصل على معلوماتنا التسويقية، ونمو الذكاء الاصطناعي، وهيكلة الإيرادات، مباشرة وبدون حشو.
اسمك مدرج في القائمة
أهلاً بكم في موجز نمو الطاقة، تابعوا بريدكم الوارد للاطلاع على النشرة القادمة.