Aleyda’s AI Search Checklist: What Changes in September 2026

A Tuesday morning, a call that changes the sorting

A client calls me on a Tuesday morning. 47,000 organic sessions per month. Catalog of 1,200 references. 14 indicators tracked in his reporting. None connected to a purchase decision.

He tells me : « We’re cited by ChatGPT Search, but sales aren’t moving. »

On screen, overall AI visibility. A single number. As if classic SEO were limited to word count.

The content held up. The measurement did not.

The cases that follow come from clients I’ve tracked over the past few months. The numbers are exact, not averages.

Every month, 15 sites pass before my eyes. Nine track impressions. Three track clicks. Only one separates citation from click.

Aleyda Solis’s checklist, updated on September 20, 2026, fixes exactly that. 12 steps. 5 presence indicators. 3 reporting layers.

This version adds a checkbox and reshuffles priorities.

Aleyda publishes it after months of field observations, in front of more than 45,000 subscribers to her SEOFOMO newsletter. It’s field work, not an opinion.

The September shift interests me more than the tool.

We’ve moved from a list of « AI best practices » to a diagnostic workflow. Journeys become measurable. Citation and click are tracked separately. And extractability is validated before producing content.

These are structural changes. They affect content teams as much as data teams.

I’ll show you the five shifts that change your priorities over the next few weeks.

Avant de rédiger quoi que ce soit, on cartographie le voyage réel de l’utilisateur. Aleyda le résume en quatre temps : on interroge, on compare, on reçoit une recommandation, puis on décide — ou pas.

Shift #1: Intent Journeys, Not Keywords

Before, we optimized to « win an AI answer. » Aleyda starts with a question : which business decisions do you want to influence?

The first step of the checklist is called « Prompts and journeys. » You no longer aim to appear for a keyword. You define intent journeys that carry economic weight.

Example. A baby care site wants to influence the decision « which car seat for a city car « . A keyword is not enough. You need a journey. The user asks three questions. Compares two models. Receives a recommendation. Clicks, or not.

Aleyda’s method requires capturing the real language of users. Not the brand’s language. Not technical SEO vocabulary.

Among my clients, 47 prompts sampled over a month reveal phrasings the team had never seen. « Stroller that fits in the trunk of a 208 ». « Headlamp that lasts 8 hours without changing batteries ».

These are decisions disguised as questions. Not keywords.

The checklist asks for a sample. Not an ideal query. A representative sample.

Forbidden : starting from Search Console data. It does not capture conversational prompts. It captures clicks on the site.

Also forbidden : internal brainstorming. The team knows its catalog. It does not know the wording of an exhausted parent at 23:47.

The right sample comes from forums, support tickets, questions on product pages, chat histories. Aleyda says it : source the real language.

Then you rank journeys by value. A « which to choose between X and Y » journey before a €600 purchase weighs more than a « definition of such-and-such term » journey.

This sorting changes editorial priorities. You stop publishing for volume. You publish for margin traction.

A bedding client doubled its click-through rate by targeting 12 precise journeys rather than multiplying generic pages. 12 journeys. Not 120.

Minimum effort produces more impact when aligned with a real decision.

First September 2026 shift : measurement starts before content. It starts in the customer’s language.

Does your next content brief start from an observed prompt or an invented query ?

Shift #2: Five Presence Indicators Instead of a Single Score

The second shift is more brutal. Aleyda asks you to measure AI presence before changing anything. With five indicators.

Not one. Five.

• Appearing in the answer.
• Being recommended.
• Receiving a link.
• Winning a comparison.
• Being described with accuracy.

The client from earlier was tracking 14 indicators. None matched this grid. He measured aggregated impressions, without knowing whether the brand was mentioned correctly or whether the link was clicked.

The checklist draws a clear line : citation and click are measured separately.

Why ? Because a brand can be cited without receiving traffic. Or receive traffic without being recommended.

In the previous edition, people chased a single « AI visibility » score. Here, the 2×2 matrix becomes the norm : strong citation / weak click, weak citation / strong click, etc.

Aleyda does not say to track the five indicators the same way. She asks you to choose one lead indicator based on the business model.

A recurring consumables e-commerce will watch recommendation. For a lead generation site, it will be the link. A niche manufacturer will target description accuracy.

A dietary supplements site gained 37 positions in ChatGPT Search in six weeks. Traffic, however, took four months to follow. If the team had only looked at clicks, they would have stopped everything.

Order of magnitude observed at a client, not a promise.

The citation / click separation is the real contribution of this edition. It avoids short-term panic.

Concretely, the checklist asks for a repeatable measurement protocol. Same prompts, same frequency, same platform, same capture time. Otherwise, variations mean nothing.

This protocol, set up at a client in September : 47 fixed prompts, recorded every Wednesday morning. In 3 weeks, we saw a 19% variation in presence. No content modified. Just the algorithm moving.

This level of variability changes everything. You stop reacting to every fluctuation. You look at eight-week trends.

What is acceptable according to Aleyda ? A stable presence score on priority journeys. Not a weekly record. She aims for durability over time. She ignores spikes.

Teams that publish an aggregated chart are mistaken. Reporting must display each indicator by page group and by role.

How many of your pages are cited without receiving a click ?

Le diagnostic, c’est la moitié du travail. Sur un catalogue de 800 références, 214 pages n’étaient pas isolables par les parseurs : un quart du catalogue, tout simplement invisible.

Shift #3: Diagnose the Gap Before Creating the 801st Page

The third shift affects the order of actions.

Before, we published. Now, we diagnose.

Step 3 of Aleyda’s checklist lists 10 readiness characteristics. Ten checkpoints that explain why a page is not retrieved or not extracted.

The reflex is to add content. The lever is to fix extractability.

Observed in September on a catalog of 800 references. 214 product pages out of 800 were not isolatable by parsers. JavaScript hid specification tables. Response engines don’t wait.

We stopped production. HTML blocks were restructured. Data came before the marketing sentence.

Result : 214 pages extracted in 21 days. Without a single additional line of content.

The checklist asks you to validate two points : is your content accessible? Is it extractable?

Accessible means crawl. Extractable means the ability to isolate the answer without template noise.

In plain terms : if an AI agent reads your page and cannot copy the size chart or the price in one action, your page remains out of reach of the answer.

Step 4 of the checklist details technical checkpoints. JavaScript rendering, semantic HTML structure, consistent structured data, no hidden content, no strict cookie dependency.

A recent case on a product page with 47 images and 12 tables. The parser only returned the title. Why ? All useful content was loaded after interaction. The crawler saw an empty page.

After correction, the page became extractable in 4 days. The number of exact citations of the product doubled in six weeks.

The checklist also requires mapping the patterns. A gap between observed presence and expected presence is not a mystery. It corresponds to one of the ten readiness points.

Then you prioritize. Aleyda suggests a scoring example. You don’t fix all 214 pages at once. You first address pages with high commercial weight.

Aleyda clarifies what is acceptable : main content is isolated without additional navigation, tables are readable in HTML, product attributes appear in the DOM.

Otherwise, the answers cite a competitor : its specification table is easier to copy.

Making existing content readable before publishing pays off. This shift goes against the production reflex. It brings clarity.

How many of your top 100 product pages are extractable today ?

Shift #4: Third-Party Source Ecosystem Comes Before Link Ratios

The September 2026 edition devotes an entire section to the third-party source ecosystem. Only one word matters here : ecosystem.

Aleyda writes it in black and white : the composition of sources matters more than the ratio.

We no longer chase the number of mentions. We think in roles.

A DIY e-commerce site, for example, needs three types of proof : reviews, independent comparisons, forum answers. Each plays a different role in the AI response.

The checklist offers a 6-dimension grid to score each citation opportunity. Source reputation, context relevance, page durability, alignment with the business model, probability of being copied by the agent, domain authority.

Before, we bought 30 links. Today, we coordinate 8 sources covering the same semantic angle.

A furniture client saw 62% of its AI citations come from two platforms. Only one changed its moderation algorithm. Visibility dropped 14% in one week. Why ? The ecosystem lacked diversity.

The checklist requires classifying each source by role : product proof, usage proof, authority proof.

You then allocate budget platform by platform. Not globally.

Aleyda reminds us that the platform mix changes the budget. All sources are not equal depending on the market. A marketplace in France does not carry the same weight as a forum in Germany.

She adds a nuance : formats to maximize and activities to avoid.

Maximize : factual answers directly citable, comparison tables, transcribed videos, detailed usage guides.

Avoid : self-promotional pages without proof, fabricated reviews, artificial community signals.

The checklist closes the door to prompt injection and citation gaming tactics. Aleyda says it : they pay off in the short term, but destroy durable trust.

This is a strong point of the September 2026 edition. We no longer optimize against systems. We build a mesh of evidence that agents cite with confidence.

Coordination of this ecosystem must be assigned. One owner per source type. Not an intern sending partnership emails.

At a beauty sector client, we reduced the number of sources from 23 to 9. We strengthened the quality of three of them. AI presence advanced by 11 points on the 47 tracked prompts, without adding a single link.

The lesson : proof density on one angle is worth more than the volume of referring domains.

Counterintuitive ? You need a coherent composition, not a volume of mentions.

How many third-party platforms cite the same product page today ?

Shift #5: Business Data Becomes Machine-Readable

The fifth shift is the least visible. It concerns transactional data.

Aleyda dedicates step 9 to it : making business and transactional information machine-readable.

For an e-commerce, that means : product feeds, merchant programs, prices, stock, lead times, reviews. Not just on the site. In structured formats that platforms can read.

I audited a site with 1,200 SKUs. The product feed was not up to date. Prices displayed in generative answers were 19 days old. Delivery details were missing.

Result : the site won answers, but lost the agent’s trust. Recommendations went to a competitor with fresher data.

The checklist details implementation checks by business model. Subscription, retail, local service. Each case has its critical fields.

For subscription : minimum duration, renewal conditions, monthly cost, cancellation terms. For retail : price, stock, delivery lead time, return cost. For local service : service area, time slots, hourly rate.

These fields are not editorial content. They are variables. Without them, the agent cannot recommend precisely.

Aleyda insists on feeds and merchant programs. An up-to-date Google Merchant Center feed is not a bonus. It’s an entry condition for purchase answers.

A catalog of 3,400 references lost 23% of its AI impressions in three days. The cause ? An undetected feed format change. Prices had disappeared from answers.

The checklist suggests implementation checkpoints. Required fields, update frequency, consistency between the site and the feed, accessibility of structured data.

And localization is no longer limited to language. Aleyda insists : localize by market, not only by translation. Currencies, standards, date formats, cultural expectations. A « 220 V compatible » product does not mean the same in Southeast Asia as in Europe.

She cites the example of differences between markets : the same headphone model sold in France and Japan will not have the same selling points. Nor the same warranty-related questions.

Localization by market requires coordination with the product team and logistics. Not just SEO.

A simple check : ask ChatGPT Search for the price of one of your best-selling products. If the answer differs from your page, you have a problem. Not a content problem. A feed problem.

Transactional data is your most important content. If it’s wrong, no semantic phrasing will save the recommendation.

Is your product feed up to date within the last 7 days ?

Reporting With Controlled Claims, and What the Checklist Does to Your Teams

The last section of Aleyda’s checklist tackles reporting. 3 layers.

• Layer 1 : presence.
• Layer 2 : readiness.
• Layer 3 : business impact, by confidence level.

We stop presenting a single number. We present a bundle of signals with confidence levels.

First-party data from AI platforms remains limited. Aleyda reminds us : we build an observable, cautious layer.

This caution is a method.

Layer 1, presence, answers the question : do we appear? Layer 2, readiness, answers : are our pages prepared? Layer 3, business impact, answers : did it change sales? With degrees of confidence.

We no longer say « +32% AI visibility ». We say : presence progression on 47 prompts, medium confidence, correlated with +4 attributed orders, low confidence.

This language changes the relationship with executives. Overclaim disappears.

Aleyda details the first-party data available by platform. Some expose clicks. Others don’t. Some give rankings. Others don’t. Reporting must adapt.

This 3-layer reporting, deployed at a furniture client in September, transformed meetings. Within four weeks, we no longer discuss impressions. We discuss hypotheses to validate.

This is where the DOSE framework comes in. As I teach it with Guillaume Attias’s BMO Academy, the first layer of a resilient organization is the feeling of mastery. A clear checklist, checked step by step, triggers completion dopamine.

Not motivation through speeches. Progression through protocol.

When an e-commerce team sees the 12 steps and can validate the first one unambiguously, they adopt the method. The validation loop becomes a routine, not a constraint.

Aleyda suggests a minimum viable workflow : cadence, hypotheses, owners. Each change must have an indicator to validate afterward.

A rhythm, not an annual audit.

She gives an example of a loop : you modify a page’s extractability. You wait two weeks. You measure presence before / after. You validate or invalidate the hypothesis. You adjust.

What matters : a loop that learns.

That’s the September 2026 shift : no one asks you to be « AI optimized ». They ask you to prove that your next action will have a measurable effect.

How many of the 12 steps does your team validate today, with controlled claims ?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn