Skip to content

How Seerly works

The loop behind every number in Seerly, from the questions asked through to the work it produces.

StartUpdated 25th August 20263 min read

Overview#

Traditional SEO tools measure where you rank in a list of links. Seerly measures something different: when someone asks an AI assistant for a recommendation in your category, does it mention you, how does it describe you, and which sources did it draw on?

There is no ranking to check. The answer is written fresh each time, and the brands named in it are chosen by the model. Seerly's job is to make that repeatable enough to measure.

The loop#

The loop

your moveSeerly runs this
  1. 1

    Category becomes topics

    Insights → Keywords → Universe

    Search demand grouped into what gets measured

  2. 2

    Topics become prompts

    Actions → Prompts

    Realistic buyer questions, per topic

  3. 3

    Prompts are asked

    The same questions, every cycle, against your tracked engines

  4. 4

    Answers are read four ways

    Insights

    Visibility, sentiment, citations, competitors

  5. 5

    Gaps become work

    Insights → Gap Analysis

    What is cited instead, and what those pages share

what you publish changes the next run's answers

The first node decides everything downstream, which is why the topic set is worth checking before any score. Only the last move is yours, and it is the one that closes the loop.

1. Your category becomes a set of topics. Seerly maps the search demand around what you sell and groups it into topics. During setup this happens for you, steered by the focus areas you entered. This is the step that determines everything else, because those topics are what gets measured. You can change the set afterwards from the Keyword Universe.

2. Topics become prompts. Each topic produces realistic buyer questions, phrased the way people actually ask AI assistants: recommendation questions, comparison questions, experience questions.

3. Prompts are asked, on a schedule. The same prompts run every cycle, against the AI engines your plan covers. Asking the same questions repeatedly is what turns an anecdote into a trend.

4. Answers are read four ways. Whether you appear at all (visibility), how you are described (sentiment), which sources the answer leaned on (citations), and who appeared instead of you (competitors).

5. Gaps become work. Where you are absent, Seerly shows the pages being cited instead and what they have in common, which turns an observation into a content brief.

The discipline this belongs to has several names, and the vocabulary is still settling. See Generative Engine Optimization for what GEO, AEO and AIO all refer to, and how it differs from search optimization.

Why the same question gives different answers#

AI answers vary between runs even when nothing whatsoever about your brand has changed. This is normal behaviour rather than a defect in the engines or in the measurement, and it is precisely why Seerly asks a fixed set of prompts repeatedly instead of asking once and reporting the result.

The practical consequence is that a single run is not evidence of anything. One appearance is not a win, one absence is not a loss, and building a plan on either will send you somewhere arbitrary. What you are looking for is whether something recurs, because recurrence is the part that survives the variation.

Best practices#

Healthy movement looks gradual and directional across several runs rather than dramatic in any one of them. A number that swings wildly from run to run is usually telling you the prompt set behind it is too small to be stable, not that your brand's standing genuinely changed overnight, and the fix for that is more prompts rather than more panic.

Next steps#

  • Judge trends rather than runs. Give any change a few cycles before you act on it, and considerably longer before you report it as a result.
  • Fix topics first if they look wrong. A topic set that does not describe your business will make every downstream number a careful description of somebody else's.
  • Treat recurrence as the signal. A source cited once is noise; a source cited every single run is the thing you are actually competing with.

Limitations#

Seerly reads public AI answers and public pages. It has no access to what any individual user was shown in their own session, and no ability to edit what a model says about anybody.

Coverage depends on your plan. See seerly.app/pricing.

Related