Method
How we turn AI-answer screenshots, tool exports, and competitor mentions into one defensible next move
A screenshot in the founder's Slack. A dashboard export no one quite trusts yet. A competitor turning up in a comparison answer your sales team did not know existed.
By now someone senior has probably asked: is this a problem, or is this one weird output?
That is the question the AnswerPresence Method is built to answer.
We build the prompt sample around your buyers' actual situations, read each segment separately, check which patterns hold across repeated runs, diagnose what is driving the gaps, and give one commercial recommendation.
Act internally. Run a diagnostic. Scope a pilot. Or leave it alone for now.
01
The method in four steps
The AnswerPresence Method is simple on purpose.
The judgement underneath is not.
- 01
Benchmark the category.
We compare your brand against 3-5 named competitors across the prompts your buyers are likely to run.
- 02
Find the lost answers.
We isolate the prompts where you're absent, weakly framed, or losing ground at a moment that matters.
- 03
Identify evidence and citation gaps.
We trace the sources AI tools are leaning on and check what your pages and wider source set give them.
- 04
Recommend the next move.
One evidence-backed decision: internal action, diagnostic, pilot, or no further work.
The rest of this page explains the discipline behind those four steps.
02
We start with buyer situations,
not generic prompts
The first job isn't running prompts.
It's working out which prompts belong in the sample.
That starts with your inputs: competitors, priority pages, positioning, known category language, and any prompts the team already cares about.
It also starts with a more direct question: has anyone important already seen a missing answer? An investor, founder, board member, sales lead, or senior marketer who went looking and found a competitor where your brand should have been?
That kind of prompt does not need to be representative to be worth including. It shows where the commercial anxiety already lives.
From those inputs, we build a prompt universe across the situations that matter:
- Branded prompts
Where the question is answer hygiene and source quality, not whether AI tools have heard of you
- Bottom-of-funnel prompts
Where the buyer knows the category and is comparing options
- Competitor and switching prompts
Where the buyer is close to a decision or looking for an alternative
- Exploratory prompts
Where the buyer has a problem but has not identified the solution category yet
- Jobs-to-be-done prompts
Where the buyer describes the work they are trying to get under control
- Industry-specific prompts
Shaped by regulations, certifications, workflows, integrations, or security requirements
A flat prompt list gets more coverage.
A prompt universe gets the right coverage.
The method is not trying to sample every possible question. It is trying to sample the questions where an AI answer could shape consideration, confidence, or the next click.
03
Different prompt types, different readings
Brand prompts and category prompts do not belong in the same score.
When someone asks about your company by name, you should probably appear. That is not what the audit is measuring. It is measuring how AI tools describe you, which sources they draw on, whether competitors have found their way into your branded answers, and whether the message matches what your team would want a buyer to hear.
That is a hygiene reading.
A category prompt is different. So is a competitor or alternative prompt. So is an exploratory prompt from a buyer who has not named a category yet.
Same dashboard.
Very different decision.
Within each segment, the audit looks at mention rate, citation rate, share of voice, answer position, competitor co-mentions, cited sources, source types, and how the answer frames the category or the brand.
Knowing where you're strong, where you're absent, and where answer quality is weak points to different actions. A single blended score obscures which one applies.
04
We look for what keeps happening,
not what happened once
A screenshot proves something happened once.
The more useful question is what keeps happening.
For the prompts that matter most, the audit looks for recurring patterns across repeated runs, different tools, and different time windows.
Not because repetition gives certainty. AI answers shift by tool, phrasing, session, and time. What repetition gives is a better read on consensus.
Which competitors keep appearing? Which sources keep getting cited? Which features, use cases, and discussion points keep showing up as important? Where does your content match that consensus, and where does it fall short?
A single answer can be unusual.
A pattern that keeps coming back is harder to wave away in an internal conversation.
A pattern that does not come back should not be funding a pilot.
05
We look at why answers win
Once the audit knows a pattern is real, it asks what the winning answer appears to be built from.
That question matters because the answer differs.
- 01
Sometimes the answer is rewarding a page type. The prompt appears to want a product page, comparison page, pricing page, or use-case page. A competitor has one. Your equivalent information exists, but it lives in a blog post, a PDF, or a sales deck.
- 02
Sometimes the answer is rewarding proof placement. It keeps pulling in security, integration, migration, customer evidence, or category proof, but your page only gestures at it. The evidence exists. It just is not visible where the answer appears to be drawing from.
- 03
Sometimes the answer is rewarding structure. A competitor page gives AI tools a clear table, answer-first section, comparison block, or sourceable sentence. Your page covers the same ground, but in prose that is harder to lift, compare, or cite.
- 04
Sometimes the framing comes from a third-party source. A review aggregator, comparison site, partner page, or industry list is doing the work, and your owned pages may not be the main surface shaping the answer.
- 05
And sometimes the answer keeps treating a discussion point as important. A workflow step. A use-case combination. A capability the buyer appears to be weighing.
It simply is not on your pages at all.
Different cause, different fix.
A page-type mismatch is not solved by writing more content. A proof-placement gap is not solved by ranking a new page. A third-party framing problem usually needs source work, not another owned-page rewrite.
The goal is not to reverse-engineer every algorithm.
It is to know which diagnosis you are dealing with, then show the evidence chain behind that read.
06
What the audit traces
A recommendation is only useful if the evidence behind it can travel.
For the prompts that matter most, the audit builds a chain: the prompt segment, the exact prompt text, the answer returned, which brands appeared and where, which sources were cited and what type they are, the gap type, how much the gap appears to matter, and the next action.
That sequence lets interpretation sit alongside the reading, not obscure it.
When the report says a competitor appears to be winning on page type or source set, you can see the prompt, the answer, and the sources behind the call. If someone needs to challenge the interpretation, the reading is right there.
Not every lost answer deserves a project. Some gaps are noise. Some are hygiene. Some point to a page or source issue worth fixing now.
The receipts file holds the prompts tested, the full answers returned, the brand and source citations, and the URLs behind the findings. You don't need to read it to act on the report. It's there for the moment someone senior asks where a specific claim came from, and "the audit said so" is not the answer you want to give.
That is what lets the finding travel without an asterisk.
No mystery layer.
No "trust us" box.
And no claim that the chain proves causality. It traces what was seen, what the audit thinks it means, and what to do next. The judgement is visible. So are its limits.
07
What the method will not pretend to be
Restraint is part of the method, not a hedge against it.
- It will not call one prompt run a benchmark. A single answer can tell you something. It cannot tell you what keeps happening.
- It will not blend branded, competitor, category, and exploratory prompts into one tidy score. Different prompt types point to different actions. A blended number hides which one applies.
- It will not treat tool data as the whole audit. If you already have an AEO tracker, that data is welcome as an input. The audit's job is to interpret it alongside repeated prompt checks and decide what is worth doing next.
- It will not pretend that AI-answer attribution, citation value, or prompt stability can be measured cleanly. The category is not mature enough for that, and false precision makes the finding harder to trust.
- It will not turn every gap into a case for a pilot. The recommendation can be act internally, wait, run a smaller diagnostic, scope a pilot, or leave the issue alone for now. The evidence decides which.
What you need is the standard of evidence, the shape of the method, the limits, and the confidence that the recommendation can travel.