This is a template, not a case study. AIgnored is new and has not run a client engagement, so publishing a story with invented numbers would be exactly the thing this product exists to argue against.
What follows is the structure to use when you run the work yourself, along with the measurement rules that decide whether the resulting document is evidence or decoration. Fill in the brackets with what actually happened, including the parts that did not work.
The template
Headline
{Business type} in {city}: named in {n} of 8 AI answers after 60 days, from 0
Lead with the metric and the timeframe. No adjectives in the headline. If the number is 3, say 3; a modest true number is more persuasive than a large vague one.
The business, in three lines
- What they do, where, and how long they have been trading
- The thing that made the invisibility surprising: rating, review count, years in business
- What they wanted: more of a specific kind of job
Keep it to three lines. The reader is not here for the client.
The baseline
State the method before the numbers, in one short paragraph:
On {date} we asked {list the assistants} eight questions a customer would ask, each in a fresh conversation from a logged-out session. We recorded every business named in every answer.
If you have not settled on a method yet, the 15-minute manual audit is one you can describe in that one paragraph.
Then the table. Client at 0. Top three competitors at whatever they were. Nothing else.
Include one screenshot of a raw answer that does not contain the client's name. The screenshot does more work than the table.
What we found in the sources
The diagnosis, in three or four bullets. This is where you show you did more than run a test:
- Which pages the answers were built from
- Which of them the client was absent from
- What was wrong on the profile (category, missing services list, hours)
- Which question shapes were winnable and which were not
Be specific. "Their primary category was Contractor rather than Plumber" is a finding. "Their online presence needed work" is filler.
What we did, week by week
A dated list. No strategy language. The plan runs to day 60, so the re-audit matches the headline.
Week 1: primary category changed from {x} to {y}. Services list written out, {n} entries. Description rewritten with {n} towns named in text. Week 2: listed on {directory}, {directory}, {association}. Name standardised from {n} variants to one across {n} properties. Weeks 3 to 4: wrote two pages, {page} and {page}, each answering one question shape in plain sentences with a price range. Week 5: changed the review request wording to ask what the job was and how fast we arrived. Weeks 5 to 7: {n} new reviews in three weeks, {n} of them mentioning response time. Week 8: no new changes, so the re-audit measures work that was already live. Day 60: re-audit with the same eight questions under the same conditions as the baseline.
The value of this section is that a reader can copy it. Let them.
Keep the quiet week. Changes can take a while to reach the pages an answer is built from, and a re-audit the day after the last change risks reporting less than the work did.
The result
Re-run the identical questions under identical conditions. Report:
- The new score, and which question shapes it came from
- What did not move, and your best explanation why
- Any competitor movement, because it contextualises yours
A result that says "we went from 0 to 4, entirely on the emergency and service-specific questions, and we are still absent from the broad comparison question" is far more credible than "we improved visibility across the board".
The caveats section
Include it. It is the section that makes the rest believable.
Three things to state plainly:
Variance. These models are not deterministic. Re-running the same questions produces different answers, so a score of 4 might read 3 or 5 next week. The finding that survives noise is the move away from 0.
Attribution. If the client's phone rang more, you probably cannot prove it was this. Say what you can support: mentions went up, branded search went up, the intake script recorded {n} people saying an assistant recommended them. Do not convert that into a revenue figure you cannot trace.
Method boundaries. If you measured through provider APIs rather than the consumer apps, say so and say why it is a close approximation rather than a transcript. Our methodology page explains that difference in full and it is worth linking rather than re-explaining.
What it cost
Publish the fee if you can, and publish the time. A case study without a cost is a story; with a cost it is an offer. The pricing guide for agencies covers how to price the sprint behind it.
The rules that make it honest
Same questions, same conditions, both times
Write the eight questions down before the baseline and never edit them. Changing a question between the before and the after invalidates the comparison, and it is easy to do by accident when a different person runs the second audit.
Same conditions: logged out, fresh conversation for every question, same assistants, search enabled.
Keep the raw answers
Save every answer, both runs. If a client, or a prospective client, ever asks to see the evidence, the raw text is the answer. A tool that gives you a score with no way to read the underlying responses cannot support a case study.
Report the failures
The question shapes that did not move belong in the document. So does the competitor who overtook you in week 4.
This feels like it weakens the case study. It does the opposite: a document with one honest failure in it reads as a measurement, and a document with no failures reads as marketing.
Do not annualise
"Went from 0 to 4 mentions in 60 days" is a fact. "On track for 24 mentions by year end" is a fabrication: it projects a rate from a single 60 day window, and one window tells you nothing about what the next five do.
Two shapes a first engagement can take
This section is a hypothesis, not a finding. We have not run these engagements, so what follows is reasoning from the mechanism, and your own before-and-after is what confirms it or kills it.
A business can be at 0 for structural reasons: wrong primary category, absent from every directory the answers cite, no services list. Those inputs are mechanical to change, which is the argument for trying them first. If changing them moves nothing, that is a finding worth publishing too.
A business can also be at 0 while already doing the basics, competing against names that come back under every phrasing. There the narrow service-specific questions should be the winnable ground rather than the broad comparison question, and the promise to make is "we will work on the two questions that convert" rather than "we will get you named".
Deciding which one you are looking at before you quote is the part that takes judgement, and the baseline audit is where you make the call: which question shapes the client is absent from, which pages the answers cite, and how much of the profile is simply missing. Whether that call predicts the result is what your first few case studies are for.
Building the pipeline behind the case study
One case study is worth writing. It is also worth remembering that the engagement started with an audit of a business that had not asked for one, and the cold email templates for AEO outreach show how to send that audit without a pitch.
That is the volume problem. Auditing one prospect by hand is about fifteen minutes with the manual method. Auditing every plumber in a city, so you can pick the ones with the fixable problem, is a week. AIgnored exists for that step: it asks its own fixed set of 8 broad recommendation questions per model for every business in a city and niche, scores them, and marks the ones with strong reviews and a real website that the assistants never name. It does not ask service-specific questions, so the shapes a case study usually reports on still need the eight questions you wrote down for the baseline. A business at 0 is the one where a change in the score is unambiguous rather than noise, which is another way of saying it is where your next case study comes from.
It runs on your own API keys, so a scan costs whatever the provider charges you. The pricing page has a free plan covering a single city, which is enough to produce one baseline audit and, sixty days later, one real case study with numbers you did not have to invent.