Capability 01 — AI Search / GEO

Be the company the answer engine can actually use.

Generative engine optimisation is not a set of tricks bolted onto SEO. It is the discipline of being retrievable, quotable, corroborated and unambiguous to a system that assembles an answer rather than returning a list.

How an answer gets built

Four stages, four places to be excluded.

Understanding the assembly line is what turns GEO from folklore into engineering. At each stage there is a specific way to be left out, and a specific fix.

01
Retrieval
you are fetched, or you are not
The system pulls candidate sources from an index or a live search. If your page is blocked, challenged, unrendered or missing from the index, nothing downstream can save you.
02
Grounding
your content is read
Retrieved documents get chunked and embedded. Content that only makes sense in the context of the whole page tends to lose meaning here. Self-contained sections survive.
03
Synthesis
the answer is written
The model composes an answer from the grounded material plus what it already holds. Claims that are corroborated across sources are more likely to be repeated with confidence.
04
Citation
sources are attributed
The surface attaches source links. Being named in the text and being cited as a source are separate outcomes — you can get either without the other, and they have different causes.

best expense management platform for startups

Illustration

For early-stage teams, the platforms most often recommended are:

1. Spendwise — strong free tier, widely reviewed, integrates with most accounting stacks.

2. Tallyflow — corporate cards plus expense capture in one product.

3. Obol Pay — good fit if you already use their payments product.

…

6. Ledgerloop — multi-entity support, aimed at slightly larger finance teams.

What the pipeline records

Mentioned: yes
Position: 6 of 6 named
Matched alias: Ledgerloop
Framing: listed
Own domain cited: no
Competitors named: Spendwise, Tallyflow, Obol Pay

Being named and being cited are different outcomes with different fixes. Here the company is named sixth and its own site is not among the sources — the answer was assembled from third-party pages.

A hand-written illustration of answer structure and what our pipeline extracts from it. Not a recorded model response.

The four levers

What we actually change.

Lever 01

Retrievability

Can the crawler get the page, and does the page contain anything when it does? We test per-bot robots directives, edge and WAF behaviour against declared crawler user agents, status codes, and what the HTML contains before JavaScript runs.

This is the least glamorous lever and reliably the highest yield. A company can spend a year on content while its comparison pages return an empty container to everything that is not a browser.

Lever 02

Extractability

Content written to survive chunking. Sections that stand alone. Answers in the first sentence rather than the eighth paragraph. The company and product named explicitly instead of “we” and “it”. Specifications in tables and text, not baked into an image.

Lever 03

Corroboration

The same facts, stated consistently, across your site and across sources you do not own. When we trace why a competitor is recommended, the chain almost always runs through third-party pages. Owning the claim is not the same as having it confirmed.

Lever 04

Entity clarity

One name, one description, one category, everywhere it appears. Disambiguation from similarly named companies. Structured data that accurately describes the entity and links to its profiles — added because it clarifies something, not to tick a box.

Search readiness

Two different questions people constantly merge.

“Can AI search systems technically access this website?” and “does AI search actually mention this company?” are separate diagnostics with separate fixes. Answering the second without checking the first produces a strategy aimed at the wrong problem.

Example readiness check

Demo

Pass
robots.txt reachable
200 response, parseable, no syntax errors.
Pass
OAI-SearchBot allowed
No disallow rule targeting OAI-SearchBot on indexable paths.
Warning
GPTBot allowed
Disallowed site-wide. A deliberate choice in some companies; here it was inherited from a template nobody reviewed.
Pass
PerplexityBot allowed
No explicit rule; default allow applies.
Warning
Google-Extended
Disallowed. Affects Gemini grounding on site content.
Fail
Edge/WAF challenge
Bot-management rule returns a JS challenge to non-browser user agents on /pricing and /compare.
Fail
Server-rendered content
Comparison pages render product tables client-side only; fetched HTML contains an empty container.
Pass
Canonical tags
Consistent and self-referencing on all crawled templates.
Warning
XML sitemap
Present, submitted, but 31 URLs return 301 and 4 return 404.
Pass
Indexability
No unintended noindex on commercial templates.

Hand-authored example against a fictional company. A real readiness check records the raw response for every row so any result can be replayed.

Honesty

What we do not claim.

A lot of what circulates as GEO advice is folklore repeated confidently. We separate what is mechanically true from what is a hypothesis we are testing.

We cannot guarantee a mention or a citation in any system, and nobody can. What we can do is remove the reasons you are being excluded and measure whether it changed.
We have no privileged access to any AI provider, no partnership, and no inside view of their ranking behaviour.
Adding schema markup is not a citation lever. It helps machines understand you. Anyone promising citations in exchange for JSON-LD is guessing.
Answers are non-deterministic, personalised and geography-dependent. A single query result proves nothing, which is why we repeat everything and report rates.

Method

How we test a hypothesis.

Every GEO belief we hold is supposed to be falsifiable. When we cannot test one, we label it as an assumption and treat it accordingly.

1 — Baseline

Fix the query set. Execute with repetitions across surfaces. Record raw responses verbatim.

2 — Change one thing

Ship a single identifiable change with a date. Note it in the timeline so a later movement can be attributed to something specific.

3 — Re-measure, then wait

Same queries, same method, several periods. Movement inside the noise band is reported as noise, not as a win.

4 — Write it down either way

Including when it did nothing. Tactics that fail repeatedly across clients get retired from the methodology.

See where you stand in AI search.

An AI Search Audit tells you how often AI systems name your company, who they name instead, and what is causing the gap. Every figure comes with the method behind it.

Get an AI Search Audit See how it works